Do Institutional Investors Have an Ace up Their Sleeves?

--Evidence from Confidential Filings of Portfolio Holdings1

Vikas Agarwal 2 Wei Jiang3 Yuehua Tang4 Baozhong Yang5

First Draft: August, 2009

This Draft: November, 2009

ABSTRACT This paper studies the holdings by institutional investors that are filed with a significant delay through

amendments to Form 13F and that are not included in the standard 13F holdings databases (the

“confidential holdings”). We find that asset management firms (hedge funds and investment

companies/advisors) in general, and institutions that actively manage large and risky portfolios in

particular, are more likely to seek confidentiality. The confidential holdings are disproportionately

associated with information-sensitive events such as mergers and acquisitions, and include stocks

subjected to greater information asymmetry. Moreover, the confidential holdings of asset management

firms exhibit superior risk-adjusted performance up to three to four months after the quarter end,

suggesting that these institutions may possess short-lived information. Our study highlights the tension

between the regulators, public, and investment managers regarding the ownership disclosure, provides

new evidence in the cross-sectional differences in the performance of institutional investors, and

highlights the limitations of the standard 13F holdings databases.

JEL Classification: G10, G19

1 The paper has benefited from comments and suggestions from Nicole Boyson, Mark Chen, Conrad Ciccotello, Meyer “Mike” Eisenberg, Merritt Fox, Jeff Gordon, Gerald Gay, Laurie Hodrick, Lixin Huang, Narasimhan Jegadeesh, Jayant Kale, Omesh Kini, Chip Ryan, and seminar and conference participants at Columbia Business School, Columbia Law School, Georgia State University, University of Buffalo, and the All-Georgia Finance Conference. The authors thank George Connaughton, Bharat Kesavan, Vyacheslav Fos, and Linlin Ma for excellent research assistance. 2 J. Mack Robinson College of Business, Georgia State University, 35 Broad Street, Suite 1207, Atlanta, GA 30303. Research Fellow at the Centre for Financial Research (CFR), University of Cologne. Tel: 404 413 7326, Email: vagarwal@gsu.edu. 3 Graduate School of Business, Columbia University, 3022 Broadway, Uris Hall 803, New York NY 10027. Tel: 212 854 9002, Email: wj2006@columbia.edu. 4 J. Mack Robinson College of Business, Georgia State University, 35 Broad Street, Suite 1221, Atlanta, GA 30303. Tel: 404 413 7313, Email: fncyttx@langate.gsu.edu. 5 J. Mack Robinson College of Business, Georgia State University, 35 Broad Street, Suite 1243, Atlanta, GA 30303. Tel: 404 413 7350, Email: bzyang@gsu.edu.

1

Do Institutional Investors Have an Ace up Their Sleeves? --Evidence from Confidential Filings of Portfolio Holdings

This paper studies the holdings by institutional investors that are filed with a significant delay through amendments to Form 13F and that are not included in the standard 13F holdings databases (the “confidential holdings”). We find that asset management firms (hedge funds and investment companies/advisors) in general, and institutions that actively manage large and risky portfolios in particular, are more likely to seek confidentiality. The confidential holdings are disproportionately associated with information-sensitive events such as mergers and acquisitions, and include stocks subjected to greater information asymmetry. Moreover, the confidential holdings of asset management firms exhibit superior risk-adjusted performance up to three to four months after the quarter end, suggesting that these institutions may possess short-lived information. Our study highlights the tension between the regulators, public, and investment managers regarding the ownership disclosure, provides new evidence in the cross-sectional differences in the performance of institutional investors, and highlights the limitations of the standard 13F holdings databases.

Mandatory disclosure of ownership in public companies by investors is an essential part of the securities

market regulation. At the core of this regulation is the Section 13(f) of the Securities Exchange Act of

1934 that requires institutional investment managers to disclose their quarterly holdings.6 The quarterly

reports, filed to the Securities and Exchange Commission (SEC) on the Form 13F, disseminate to the

public information about holdings and investment activities of institutional investors. An exception to the

rule, however, provides confidential treatment of certain holdings through amendments to the original

Form 13F. With adequate written factual support, this provision allows the institutions to delay the

disclosure of some of their holdings, usually up to one year. Throughout the paper, we refer to these

amendments as “confidential filings,” and the positions included in such filings as “confidential holdings.”

A large literature has evolved based on the reported quarterly portfolio holdings of institutional

investors to evaluate these investors’ return performance and managerial ability, to extract information

from the reported holdings to form investment strategies, or to relate institutional ownership to corporate

policies and events. However, the prior papers use only the data on original 13F filings, usually from the

Thomson Reuters Ownership Data (formerly the CDA/Spectrum database), and therefore ignore the

confidential holdings because they are usually not included in the standard commercial databases.

6 Section I.A. contains a more detailed description of the institutional background regarding the ownership disclosure.

2

Apart from minimizing price impact during ongoing acquisitions and dispositions, incentives to

seek confidentiality most likely arise from private information as perceived by the investment manager. It

is in the best interest of investment managers not to disclose their informed positions before they have

reaped the full benefits of their superior information. Such incentives are often in conflict with the

regulatory rules. For example, Perry Corp, a well-known hedge fund, attempted to keep secret its

accumulation of position in Mylan Inc. in 2004 when the company was contemplating a merger with King

Pharmaceuticals Inc. The deal ultimately fell through; nevertheless, Perry was under investigation by the

SEC on the allegation of improperly withholding details about a large investment in an effort to profit. 7

Though the two parties settled in July 2009, the case highlights continuing tension between the desire of

some investors to withhold information that could reveal their investment strategy, and the demand of the

public and regulators for transparency.

As a matter of fact, several hedge funds and successful investors including Warren Buffett have

appealed to the SEC for an exemption from revealing their positions in the 13F forms but have been

unsuccessful in convincing the SEC. Philip Goldstein, an activist hedge fund manager at Bulldog

Investors likens his stock holdings to “trade secrets” as much as the protected formula used to make Coke,

and contends that complying with the 13F rule “constitute[s] a ‘taking’ of [the fund’s] property without

just compensation in violation of the Fifth Amendment to the Constitution.”8 In the wake of the “quant

meltdown” in August 2007, quant hedge funds blamed the ownership disclosure for inviting “copycats”

into an increasingly correlated and crowded space of quant strategies, which contributed to the “death

spiral” in the summer of 2007 when many funds employing similar strategies attempted to cut their risks

simultaneously in response to their losses (Khandani and Lo (2007)). Most vocal among them was D. E.

Shaw & Company who demanded confidentiality for its whole portfolio in order to guard its proprietary

models, but the request was denied by the SEC.

7 For the SEC litigation release of this case, please see: http://www.sec.gov/litigation/admin/2009/34-60351.pdf. Perry was accused of violating the rule regarding Schedule 13D which requires prompt and proper disclosure of positions above 5%. 8 For a more detailed discussion, see Philip Goldstein’s interview in September 12, 2006 issue of Business Week: http://www.businessweek.com/print/investor/content/sep2006/pi20060913_356291.htm.

3

Though confidential filing is meant to be the exception rather than the rule, some institutional

investors seem to have taken advantage of it for the benefit of delayed disclosure. Our study is based on a

comprehensive collection of all original and amendments to 13F filings by all institutions during the

period of 1999-2007, where the amendments include both approved and rejected applications for

confidential treatment. We find that 233 institutions (7.2% of all 13F filing institutions) have resorted to

confidential filing at least once, and the average (median) value of the confidential holdings amounts to

27.3% (12.3%) of the total value of securities filed in both the original and confidential 13F holdings.

Analyzing the complete holdings uncovers several interesting results that are consistent with

motives regarding both withholding private information and minimizing price impact. First, we find that

hedge funds and investment companies/advisors are more likely to seek confidentiality compared to

banks, insurance companies, and other institutions. Moreover, institutions resorting to confidential

treatment tend to manage large and concentrated portfolios, and adopt non-standard investment strategies

(in terms of low loadings on the common factors and high idiosyncratic risks). Second, acquisition-related

confidential holdings are more likely to consist of stocks associated with information-sensitive events

such as mergers and acquisitions, and to include stocks subject to greater information asymmetry as

measured by market capitalization, trading liquidity, analyst following, and probability of distress; while

disposition-related confidential holdings are most prominently characterized by relatively poor past return

performance. Finally, confidential holdings of investment companies/advisors exhibit higher, but short-

lived, abnormal performance compared to their own original filings.

Our study provides new evidence on the skill of asset-management firms and their ability to

benefit from their private information through confidential holdings. It also has implications for

researchers and regulators concerned with the transparency of financial institutions (especially the lightly-

regulated hedge funds and private funds) and the role of mandatory disclosure of their investments. A

thorough study based on a complete collection of institutional investors’ quarterly holdings could help

settle the controversy regarding the value and effect of the “non-transparent” holdings and identify key

factors that cause the cross sectional variation in the confidential filing activities. Finally, our study

assesses the limitations of using the conventional institutional quarterly holdings databases that mostly

4

exclude confidential holdings. While any error due to the omission in evaluating the aggregate portfolio

performance of institutions is likely to be small, there could be significant bias in analyzing positions

changes of specific types of institutions or around specific events (such as M&A and block building).

Our paper is most closely related to the literature that evaluates the performance and information

content of holdings by institutional investors. For example, Grinblatt and Titman (1989, 1993), Grinblatt,

Titman, and Wermers (1995), Daniel, Grinblatt, Titman, and Wermers (1997), Chen, Jegadeesh, and

Wermers (2000), Wermers (2000, 2003, 2006), Kacperczyk, Sialm, and Zheng (2005, 2008), Wermers,

Yao, and Zhao (2007), and Huang and Kale (2009) analyze whether mutual funds outperform their

benchmarks using the holdings data. Griffin and Xu (2009) and Aragon and Martin (2009)9 conduct a

similar analysis with another class of active managers—hedge funds. By incorporating the confidential

holdings and comparing them to the disclosed holdings, our study provides a more complete picture of the

ability and performance of a wide range of institutions.

Our paper also contributes to a strand of literature that studies the effects of portfolio disclosure

on the investment decisions of money managers (Musto (1997, 1999)), theoretical implications of

portfolio disclosure for portfolio selection and performance evaluation of mutual funds (Kempf and

Kreuzberg (2004)), the consequences of frequent portfolio disclosure such as free riding and front running

by other market participants (Wermers (2001), and Frank, Poterba, Shackelford, and Shoven (2004)), and

determinants of portfolio disclosure and its effect on performance and flows (Ge and Zheng (2006)). The

findings in our study suggest that seeking confidential treatment is one effective way for the investment

managers to attenuate some of the concerns analyzed in these papers.

The remainder of the paper is organized as follows. Section I provides background information

regarding the SEC ownership disclosure rules. Section II describes the construction of sample, presents

the overview of original and confidential filings, and outlines the empirical motivations. Section III

analyzes the determinants of confidential filings at the institution level and confidential holdings at the

stock level. Section IV presents abnormal returns of confidential holdings and their cross-sectional

9 Aragon and Martin (2009) is among the very few papers that use the original 13F filings directly, instead of the filings complied by Thomson Reuters. They examine a random sample of 300 hedge funds from the SEC EDGAR database, and do not account for confidential filings in the 13F amendments filed separately.

5

variations. Finally, Section V discusses policy implications before concluding.

I. Institutional Background

The current ownership disclosure rules mandated by the SEC consists of five overlapping parts:

Schedule 13D for large (above 5%) active shareholders, Schedule 13G for large passive shareholders;

Form 13F for general institutional holdings; Section 16 regarding ownership by insiders; and Form N-

CSR for quarterly disclosure of holdings required for mutual funds.10

Among the five regimes, the Form 13F requirement covers by far the largest number of

institutional investors: all institutions that have investment discretion over $100 million or more in

Section 13(f) securities (mostly publicly traded equity; but also include convertible bonds, and some

options) are required to disclose their quarter-end holdings in these securities. We call the date when the

Form 13F is filed with the SEC the “filing date,” and the quarter-end date on which the portfolio is being

disclosed the “quarter-end portfolio date.” According to the SEC rule, the maximum lag between the two

dates is 45 calendar days. As an exception to the rule, the SEC allows for the confidential treatment of

certain portfolio holdings of institutions for which they can file 13F amendments. The provision allows

the institutions to delay the disclosure of their holdings up to one year from the date required for the

original 13F form. This one-year period can be extended further if an instruction with additional factual

support is filed 14 calendar days in advance of the expiration date.

Gaining confidential treatment is not meant to be a trivial task and is not guaranteed. 11 The

applying institution must provide a sufficient factual basis and a statement on the grounds of the objection

to public disclosure, including a detailed description of the manager’s investment strategy, e.g., risk

arbitrage that warrants confidential treatment, along with supporting analysis. Furthermore, the evidence

for confidential treatment will not be applied to an entire portfolio appearing on a 13F, but rather on a

position-by-position basis. Finally, such applications are subject to SEC approval. The time that SEC

10 The SEC adopted enhanced rules on mutual funds expense and portfolio disclosure in 2004, requiring registered management investment companies to file their complete portfolio holdings with the Commission on a quarterly basis, instead of on a semi-annual basis as previously mandated. 11 The SEC official guideline for 13F amendments is available at: http://www.sec.gov/about/forms/form13f.pdf. Section “Instructions for Confidential Treatment Requests” details the requirements.

6

takes to review individual applications and make the decision varies, with the typical range being two to

twelve months. If denied, the institution is obligated to file an amendment disclosing all the confidential

positions immediately (within six business days from the date of denial).12

In 1998, the SEC tightened the rules and restricted the conditions for accepting the applications

for confidentiality.13 The triggering event was the confusion over the 13F reporting of investor Warren

Buffett which caused a significant decline in the share price of Wells Fargo & Co. in August 1997. The

13F form did not show Berkshire Hathaway’s well-known 8% stake in the bank, only because it was

reported in a confidential filing. But the misunderstanding in the market caused Wells Fargo’s stock price

to drop by 5.8% in one hour after Buffett’s 13F filing.14 Our sample period (1999-2007) falls into the

new regime when there are more stringent rules for 13F amendments as the applying institutions need to

convince the SEC that revelation of these holdings can hurt their competitive position.

The extreme case of D. E. Shaw illustrates the tension arising from such a process. On August 14,

2007, D.E. Shaw & Company, one of the largest quant-oriented hedge fund managers, filed an entirely

blank Form 13F for its second-quarter portfolio. That is, the fund manager was seeking from the SEC a

confidential treatment of its entire portfolio, based on the argument that “copycat investors” were

mimicking its strategies. The SEC denied the request on October 19, forcing the firm to file an amended

June 30 Form 13F on October 29. That amended filing covered 3,991 positions valued at $79 billion.

Similar but less extreme requests from D. E. Shaw were rejected by the SEC before. 15 Other frequent

users of confidential filings include hedge funds (e.g. Dolphin Asset, Stark Investments, and Magnetar

Financial) and investment bank trading desks (e.g. Lehman Brothers, Goldman Sachs & Co, and UBS).

12 For example, see http://www.sec.gov/rules/other/34-52134.pdf for the rejection of the request from a hedge fund, Two Sigma. There are several other cases of rejections of confidential treatment requests including those by Warren Buffett: http://www.sec.gov/rules/other/34-50206.htm, http://www.sec.gov/rules/other/34-43142.htm, and http://www.sec.gov/litigation/admin/34-43909.htm. 13 See http://www.sec.gov/divisions/investment/guidance/13fpt2.htm for the letter issued by the SEC in June 1998 where they explain the specific requirements and conditions for granting confidentiality. 14 For a full story, please see “Large Investors Face Stiff Rules on SEC Filings,” by Paul Beckett, The Wall Street Journal, June 19, 1998. 15 See “SEC: D.E. Shaw Disclosure Request Part of Regular Process,” by Marietta Cauchi, Dow Jones Newswires, January 2005.

7

It is worth noting that the confidential treatment under Section 13(f) does not over-ride other SEC

ownership disclosure rules. For example, a beneficial owner of more than 5% of a company’s equity will

need to file Schedule 13D or 13G, even if the position is under confidential treatment in the owner’s 13F

filing. The same can be said about the holdings disclosure required for registered investment companies

(mostly mutual funds), which was changed from a semi-annual to a quarterly basis (at a 60-day delay) in

2004. Nevertheless, there are more than sporadic observations in our sample where the confidential

position exceeds 5% (such as the Warren Buffett position in Wells Fargo) or where the filer is a mutual

fund management company (such as Price T. Row or American Funds). In such cases, the confidential

treatment may still afford the institutions effective delay if the 13F disclosure is the most binding

compared to the normal delay allowed by the Schedule 13G (45 days from the year-end) or by the

disclosure rules for mutual funds (semi-annual for most of our sample period16).

If investment managers choose to file 13F amendments for securities about which they think that

they have superior private information, these holdings are likely to be more informative than the

regularly-disclosed holdings. Despite their potential importance, confidential holdings are usually not

included in the conventional databases of institutional quarterly holdings.17 For example, the manual for

Thomson Reuters Ownership Data (formerly the CDA/Spectrum database), available through WRDS,

provides the following caveat about its S12 (for mutual funds) and S34 (for institutions) data: “The

holdings in the S12 and S34 sets are rarely the entire equity holdings of the manager or fund. There are

minimum size requirements and confidentiality qualifications.”

An example from the top 20 confidential filers illustrates the omission by the Thomson Reuters

database. The chosen institution is Stark Onshore master fund (manager number 10375 in Thomson

Reuters). In Table I, we list all the institution’s confidential holdings during our sample period, and cross 16 Obviously the confidential treatment has become essentially unnecessary for mutual funds after 2004. In fact, some mutual fund companies, such as the Capital Research and Management Company (the management company of American Funds), have requested the SEC to extend the confidential treatment to mutual fund quarterly holdings disclosure shortly after the rule change (but without success). In our sample, confidential holdings by mutual fund management companies after 2004 most likely belong to these institutions’ non-mutual fund assets. For this reason, confidential holdings cannot explain the “return gap” documented by Kacperczyk, Sialm, and Zheng (2008) after 2004. Moreover, because 13F holdings reflect the aggregate positions at the institution level, we cannot attribute confidential holdings in 13F filings of an institution to individual funds within. 17 The other potential exclusion by these databases concerns non-equity holdings, such as convertible bonds and options, see Aragon and Martin (2009) for a detailed description of this issue.

8

check with its holdings reported in Thomson Reuters. We observe that, except for one stock (Rouse Co.,

CUSIP = 77927310), all the other 54 confidential holdings in the amendments are not included in the

latter.

[Insert Table I here.]

Therefore, arguably the most interesting facet of portfolio disclosure has been this far ignored in

the extant literature. Our study fills this gap in the literature by examining the motives for seeking

confidentiality — more specifically to determine if it is information-driven, and then to estimate the

abnormal performance of the confidential holdings as well as analyzing the cross-sectional variation in

the frequency of resorting to confidentiality by and the performance of different types of institutional

investors.

II. Sample Overview and Empirical Motivation

A. Sample of Original and Amendments to 13F Filings

A key data component to this study is the original and amendments to 13F filings by all

institutions. As we mentioned in the previous section, the standard databases do not provide a complete

collection of these filings. Instead, we retrieve directly both the original and amendment 13F filings dated

between March 1999 and June 2007 from the SEC’s website (EDGAR). We start in 1999 so that our full

sample period falls into a uniform policy regime after the SEC tightened up the rules for approving

confidential treatment in 1998 (see Section I for more information); we end the sample of filings in June

2007 to allow a one-year period in ex post performance evaluation.

Despite the large variation in reporting style and format, we are able to process the complete

holdings information for 91% of all the 13F filings using automated computer programs, combined with

some manual processing. The resulting list of filing institutions in our initial sample amount to 3,315,

including 86.1% of the institutions that report their original 13F filings to Thomson Reuters over the

sample period,18 plus 174 more institutions that do not appear in the Thomson database at all.

18 We restrict our sample of original filings to the processed 13F filings directly retrieved from the SEC, rather than all holdings reported to Thomson Reuters. The idea is to maintain symmetry and comparability between original and

9

Amendments to 13F filings contain two types of information: disclosure of an increase in a

position that was previously filed in or a new holding that was previously excluded from the original

filings.19 We define a confidential holding as one that was excluded from the original filing or the

difference between the restated position and the originally filed position. Our results are qualitatively

similar if we impose a threshold for the difference in the second component or simply exclude the second

component. Based on these criteria, our initial sample consists of 1,958 confidential filings and 55,185

original 13F filings. As we discussed earlier, the amendment filings in our sample include both approved

and denied applications by the SEC. By searching for key words on the first page of the amendments,20

we are able to separate amendments filed before or upon the expiration of approved confidential treatment

and those filed in response to SEC denials. Based on this algorithm, the SEC rejected about 16% of all

the confidential treatment applications during our sample period.

Figure 1 plots the time series of both types of filings, as well as the number of approved

confidential filings, at the quarterly frequency. While the number of original filings increased steadily

over our sample period, the time series for the number of confidential filings was choppier but stays

roughly in proportion to the first series. Moreover, the SEC’s approval rate has trended higher since

2005.

[Insert Figure 1 here.]

Table II summarizes the cross-sectional distribution of both types of filings. Panel A reports the

delay in days between filings and their corresponding quarter-end portfolio dates. Over 86% of original

filings are filed within 45 days of the end of quarter, conforming to the requirement by SEC.21 On the

other hand, more than 93% of confidential filings are filed more than 45 days from the quarter-end confidential filings as the latter mostly do not make their way to Thomson Reuters. 19 There are a small number of cases where an amendment filing restates a smaller position compared to the position on the same stock stated in the corresponding original filing. We do not find any general patterns in this particular subsample. For example, such positions are not disproportionately related to acquisition or disposition motives and do not exhibit any significant characteristics at the institution or stock levels. Moreover, there is no obvious rule to impute returns on these “negative” incremental positions. Therefore, our analyses exclude these minority cases, which are more likely to represent restatement of shares due to reasons other than confidential treatment. 20 We search for the appearance of certain negative phrases on the front page to identify amendments filed in response to SEC denials of the confidential treatment such as “denied” and “no longer warranted”. 21 Aragon and Martin (2008) also found significant proportions of delayed original 13F filings (i.e., beyond 45-days). There is no explicit SEC rule regarding the penalty of occasional lateness in 13F filing, and we do not find a significant number of institutions in our sample which were repeatedly late in their original 13F filings.

10

portfolio date, justifying resorting to the amendments for delayed disclosure. Surprisingly, the distribution

of the duration does not differ qualitatively between amendment filings that result from SEC approvals of

confidential treatment and those from rejections (not tabulated). The lack of a difference has two

implications: First, some institutions may file amendments before the approved term for confidential

treatment expires (usually for a year) because the information on the confidential holdings has already

become stale. Second, even denied applications effectively afford significant delays in disclosure of

confidential holdings (the modal delay time is between six and twelve months).22

[Insert Table II here.]

In our later analysis, we exclude confidential holdings filed within 45 days of delay, as motives to

conceal positions in these filings cannot be justified. We also filter out both types of filings with

extremely long delays from their quarter-end portfolio dates: more than a 180-day delay for the original

filings and more than a 1505-day (four years plus the 45 days allowed for the original 13F filings) delay

for the confidential filings. We suspect that these observations are results of data recording errors or

irregular circumstances. These three filters combined remove less than 1.3% of original filings and about

8.6% of confidential filings (see Table II Panel A). Our results are not sensitive to the particular

numerical choices employed in these filters. Moreover, our study focuses on equity holdings for which

security-level characteristics are readily available and risk-adjusted performance measures are well-

defined.

Our final sample consists of 54,154 original filings by 3,246 institutions, and 1,592 confidential

filings by 233 institutions. Panel B of Table II summarizes the number of filings, number of institutions,

the dollar value, and the number of stocks in this final sample. Conditional on an institution files both an

original and an amended 13F for its holdings at a given quarter end, the dollar value of the stock positions

included in the confidential filings is significant: the average (median) value of securities in a confidential

filing is 27.3% (12.3%) of the value of the complete portfolio of the institution. In other words, the typical

22 The long time it takes for the SEC to reject applications for confidential treatment that lack adequate factual support could potentially invite abuse, that is, some institutions without legitimate reasons may still resort to frequent applications for confidentiality just to enjoy the effective delay in disclosure. Our informal conservation with the SEC staff indicates that institutions which received repeated rejections could receive warnings from the SEC and will be subject to more timely review in future applications.

11

institution tends to “mask” one-eighth to one-quarter of its portfolio from the regular disclosure when

they have both types of filings in a given quarter. The weight of confidential holdings in total portfolios

in terms of number of stocks is smaller, indicating that these holdings tend to be larger-than-usual

positions. The average confidential holding represents 1.2% of all the shares outstanding by the issuer, as

compared to the average position of 0.69% in the original holdings.

Finally, Panel C of Table II lists the ten institutions that were the most frequent confidential filers

during our sample period, and the ten institutions that received the highest number of rejections from the

SEC for their confidentiality applications. Berkshire Hathaway is on both lists. D. E. Shaw and Caxton

Corporation (currently renamed “Caxton Associates”), two of the top ten hedge fund companies in the

U.S. as of 2007, had been rejected by the SEC for 100% of their applications during our sample period.23

B. Motivations for Empirical Analyses

Our study is the first to present the prevalence and distribution of confidential filings by

institutions. It is natural to ask questions about the incentives and consequences of seeking confidential

treatment.

First, we hypothesize on the institutional characteristics that are associated with their propensity

to use confidential filings. Institutions that engage in active portfolio management should be the most

likely applicants for confidential treatment if the incentive is to delay disclosing holdings that could

reveal their trading strategies. The degree to which institutional investors collect and process information

and their ability to benefit from such activities will vary. However, in general, asset-management

companies and investment advisors are more likely to engage in proprietary trading strategies, where

private information is essential in delivering superior returns, compared to banks, insurance companies

and financial arms of corporations. Beyond the broad categories, certain characteristics that are indicative

23 We followed these two institutions out of the sample period. Caxton ceased to seek confidential treatment after October 2005 when eight of its applications were rejected all at once. D. E. Shaw stopped confidential filing after its last one in our sample in June 2007 for about a year. It has filed three applications since June 2008 each of which covers 2-3 stocks only (while the number was in hundreds and thousands before). All the three applications received speedy reviews and were approved by the SEC. These two cases provide some evidence about the possible SEC actions against institutions suspicious of abusing of the rules for 13F amendment filings, consistent with the discussion in footnote #22.

12

of institutions’ active management strategies—such as portfolio concentration, turnover rate, and

portfolio idiosyncratic risk—should also be associated with higher frequencies of confidential filings.

Our second strand of analysis examines the determinants of confidential holdings in terms of

stock characteristics. If the primary motivation of seeking confidentiality is to preserve the value of

private information, confidential holdings should demonstrate such features. First, confidential holdings

should consist disproportionately of stocks that are involved in information-sensitive events. For example,

an explicit case identified by the SEC where positions are allowed confidentiality is related to open risk

arbitrage positions. Hence, we expect target companies in M&A transactions to appear in confidential

holdings with higher than usual probabilities. A more general determinant for a stock to be incorporated

in the confidential portfolios of institutions is the degree of information asymmetry. Greater information

asymmetry provides more opportunities for profitable private information acquisition activities.24 This

incentivizes the institutions to conceal the positions in such stocks through confidential filings. As a result,

several proxies for firm-specific drivers of information asymmetry, such as firm size, distress risk, and

analyst following, should be correlated with the probabilities that individual stocks become part of the

confidential holdings. Another motive for seeking confidential treatment is to minimize the price impact

during an ongoing acquisition or disposition. Thus, low trading liquidity should increase the probability

of a stock being treated confidentially. Needless to say, information asymmetry and illiquidity are closely

related.

Finally, the motives to seek confidential treatment have implications on the return performance of

the confidential holdings. If some institutions resort to confidential treatment to hide their ongoing

acquisitions/dispositions that were driven by private information, or to hide a position where superior

information has yet to run its full course, then one would expect that the confidential holdings exhibit

positive abnormal performance compared to the holdings disclosed in the corresponding original filings.

And this relation should be more significant for asset management companies or institutions with active

portfolio management strategies. On the other hand, if confidential treatment is sought to minimize price

24 In fact, some investors having private information can lead to greater information asymmetry. Since we do not examine the determinants of information asymmetry, we are not concerned about the direction of causality here.

13

impact or to avoid front-running on unfinished transactions that are liquidity-driven, then the realized

performance of the confidential holdings should be neutral.25 Lastly, if the motive of seeking confidential

treatment is to make it more difficult for observers of 13F holdings (potential copycats) to reverse-

engineer the trading strategy, then the confidential holdings could be a portion of the portfolio that is

potentially informative of the trading strategy, rather than most predictive of future returns. As such, the

confidential holdings do not necessarily enjoy superior performance relative to the disclosed part of the

portfolio. Nevertheless, if such trading strategies are indeed based on superior private information, then

the performance of the overall portfolios of these institutions should be higher.

III. Determinants of Confidential Filings and Confidential Holdings

This section discusses the determinants of confidential filings (at the institution-quarter level) and

confidential holdings (at the institution-holding-quarter level). Unless otherwise specified, we incorporate

quarterly fixed effects, adjust standard errors for heteroskedasticity and clustering at the institution level,

and use the 5% level as the criterion for statistical significance.

A. Institutional Characteristics and Propensity of Confidential Filings

We resort to the following models to relate the characteristics of institutions to their propensity to

use confidential filings. The first is a probit model:

, , ,( 0) ( 0),j q j q q j qCF InstCharβ λ ε> = + + > (1)

and the second is tobit model:

*, , ,

*, ,

( ) ,

( ) max( ,0).j q j q q j q

j q j q

Latent CF InstChar

Observed CF CF

γ λ ω= + +

= (2)

The dependent variable in (1), (CFj,q > 0), is the indicator variable for the existence of a confidential filing

in the (j, q) institution-quarter. The dependent variable in (2) is the dollar value proportion of confidential

25 Such a result does not necessarily negate the benefits of confidential treatment to the institution. The institution might otherwise have to incur higher transaction costs or less favourable price impact in order to finish its acquisition/disposition. Such costs are not captured by returns imputed from changes in holdings which ignore transaction costs.

14

holdings in the total portfolio (that include both confidential holdings and holdings disclosed in the

original 13F filings) of the given institution-quarter. The regressors in both models include a vector of

institutional characteristics variables (InstChar) and quarterly dummies to control for unspecified time

effects.

Results are reported in Table III. In addition to the coefficients and their associated t-statistics,

we also report the average partial effects (APE) to facilitate the interpretation of the economic magnitude.

For the probit model, the APE is defined as:

( ), , ,Pr( 0) / | , .j q j q j q qAPE E CF InstChar InstChar λ= ∂ > ∂ (3)

Our estimates of the APE are the empirical analogue to the expression above:

( ),,

1ˆ ˆ ˆj q q

j q

APE InstCharn

β φ β λ⎡ ⎤

= +⎢ ⎥⎣ ⎦∑ , (4)

where is the standard normal probability density function. The APE associated with a covariate is

determined by both the underlying sensitivity of confidentiality-seeking propensity to this covariate (β)

and the sample distribution of all covariates (the sample average of ( )φ ⋅ ).

[Insert Table III here.]

The γ estimate in the tobit model indicates the partial effect of the regressors on the latent

variable: *, ,/j q j qCF InstChar∂ ∂ , which is not usually of interest. Instead, the more meaningful APE

concerns the effect of the regressors on the actual choice of confidential holdings, that is,

, ,/j q j qCF InstChar∂ ∂ , which could be expressed as follows:

, ,,

,

| , ,j q j q qj q q

j q

CF InstCharAPE E InstChar E

InstChar ϖ

γ λλ γ

σ

⎛ ⎞ ⎡ ⎤∂ +⎛ ⎞= = Φ⎜ ⎟ ⎢ ⎥⎜ ⎟⎜ ⎟∂ ⎢ ⎥⎝ ⎠⎣ ⎦⎝ ⎠

(5)

where ( )Φ i is the cumulative probability function of the standard normal distribution. The empirical

analogue to (5) is readily available:

,

,

ˆ1ˆˆ

j q q

j q

InstCharAPE

n ϖ

γ λγ

σ

⎛ ⎞+⎜ ⎟= Φ⎜ ⎟⎝ ⎠

∑ (6)

( )φ i

15

Table III uses two sets of InstChar variables. The first set includes dummy variable for the broad

type of institutions, namely banks and insurance companies, hedge funds, investment companies/advisors,

and others. The group of banks and insurance companies serves as the omitted category in the

regressions.

It is worth noting that the classification of institution types employed in this paper is quite

different from that used in the Thomson Reuters database. Thomson Reuters divides all institutions into

five types: banks (type code = 1, narrowly defined as financial institutions that accept and manage

deposits and make loans, or loosely “commercial banks”), insurance companies (type code = 2),

investment companies (type code = 3, mostly mutual fund management companies), independent

investment advisors (type code = 4, including asset management companies, investment banks, brokers,

private wealth management companies, etc.), and others (type code = 5, including pension funds,

endowment funds, most of the hedge funds, financial arms of corporations, and others). The type code 5,

especially since 1998, is known to be problematic in that the category could include many misclassified

institutions that should be assigned with the other type codes (mostly, type code 4).26 As a result, the

“other” category, instead of being a residual claimant, turns out to be the largest category in the Thomson

database, accounting for over 50% of all institutions.

We made the following changes to the Thomson classification of institutional categories. We

first divide all institutions into four groups: (i) banks and insurance companies (a combination of type 1

and type 2 institutions by the Thomson classification), (ii) hedge funds (the classification of which will

follow), (iii) investment companies and investment advisors (a combination of type 3 and type 4

institutions by the Thomson classification, excluding hedge funds) and (iv) other institutions. The

categories of hedge funds and investment management companies/advisors will be the subjects of our key

interests. For institutions in our sample that are not covered by Thomson, we manually classify them.

Next, we made major corrections for the “other” category as classified by Thomson. First, we

reassign all hedge funds from this category. Second, we reassign an institution which has type code 5

after 1997 to an earlier code, if available and if different from 5. Third, we manually classify the 26 The data manual of the Thomson Reuters database acknowledges this issue with the classification.

16

remaining institutions (mainly based on information from the institutions’ websites), and reassign all

investment companies and advisors. After all these corrections, the “other” category shrinks sharply to

about 4% of all institutions in our sample.

Hedge funds are classified by manually identifying hedge fund management companies from all

13F-filing institutions. A hedge fund management company is defined as an institutional investor that has

a major hedge fund management business according to the information revealed from a wide range of

sources, including the institution’s own websites and SEC filings, industry directories and publications,

and news article searches. The full list of 13F-filing hedge funds is obtained from Agarwal, Fos, and

Jiang (2009) which provides a detailed description. There are 914 unique hedge funds in our sample of

which 106 have engaged in confidential filing, making hedge funds the leading group of confidential

filers in our sample. Due to our top-down classification approach, our list of 13F filing hedge funds

companies is considerably longer than used in prior literature.27

The second set of InstChar variables consist of continuous variables mostly constructed based on

13F quarterly holdings. These variables capture the degree of active portfolio management or the market

impact of the institutions. More specifically, Age is the number of years since the institution’s first

appearance on Thomson Reuters. PortSize is the total equity portfolio size of an institution calculated as

the market value of its quarter-end holdings. Turnover is the inter-quarter portfolio turnover rate,

calculated as the lesser of purchases and sales divided by the average portfolio size of the last and the

current quarter. 28 PortHHI is the Herfindahl index of the portfolio, calculated from the market value of

each component stock. PortRet and PortVol are the monthly average return and volatility on the portfolio

during the quarter, assuming that the institution maintains the holdings of the last quarter-end. Flow is

defined as the increase in total portfolio value between two consecutive quarters, net of the increase due

27 Relying on a one-sided match from published hedge fund lists to the 13F database, Brunnermeier and Nagel (2004) focus on the holdings of 53 hedge fund companies, and Griffin and Xu’s (2009) sample contain 306 such firms. 28 Purchases (sales) are calculated as the sum of the products of positive (negative) changes in the number of shares in the holdings from the previous to the current quarter-end and the average of the stocks prices at the two quarter-ends. The logic of using the lesser (rather than the average) of purchases and sales is to free the measure from the impact of net flows—a practice used in mutual fund research (e.g., by Morningstar) in defining portfolio turnover rates.

17

to returns, and expressed as a percentage of the portfolio size at the previous quarter-end. That is,

, , 1 ,,

, 1

(1 ).j q j q j q

j qj q

PortSize PortSize PortRetFlow

PortSize−

−

− += (7)

Flow measures the increase in the value an institution’s equity portfolio due to expansion in investment

(and not due to appreciation of the stock prices), and is a proxy for the fund inflows that the institution

receives.

Finally BetaMkt, BetaSMB, BetaHML, and BetaMom are the loadings on the Fama-French three

factors (market, size, book-to-market) and the momentum factor using imputed monthly returns for the

36-month period ending in the current quarter, assuming that the institution always maintains the most

recent past quarter-end holdings.

The Columns (1) and (3) of Table III indicates that hedge funds are the most frequent users of

confidential filings. Compared to the omitted category of banks/insurance companies, the probability that

a hedge fund seeks confidentiality is 3.0 percentage points higher, and the proportion of confidential

holdings in the total portfolio is 1.5 percentage points higher. Both figures are statistically and

economically significant (relative to the all-sample average probability of 1.9% and unconditional

average proportion of 0.82%). Investment companies and advisors are also frequent confidential filers,

but their marginal effects are less than half of those of the hedge funds.

Columns (2) and (4) relate confidential filing to a more detailed set of institutional characteristics

variables. We find that the following characteristics are significantly associated with more frequent

confidential filings. First is portfolio size (PortSize), consistent with larger institutions bearing higher

market impact and maybe having larger capacity in collecting private information. Second, several

characteristics proxying for active portfolio management are uniformly associated with more confidential

filings. They include high portfolio turnover rate (Turnover), high portfolio concentration as measured by

the Herfindahl index (PortHHI), and high portfolio return volatility (PortVol).

Interestingly, institutions that seek confidential treatment more often and that have larger portion

of their portfolio masked from their original filings are the ones whose equity portfolios exhibit

significantly lower loadings on the market (BetaMkt) and the momentum (BetaMom) factors. Combining

18

higher portfolio return volatility and lower factor loadings, we can conclude that these institutions manage

portfolios that have higher idiosyncratic risk (or low R-squared with respect to the market and common

factors). We argue that such a pattern is supportive of private information and active portfolio

management. First, a recent paper by Titman and Tiu (2009) find that better hedge funds (in terms of

Sharpe ratios, information ratios, and fund inflows) exhibit lower R-squared values with respect to

systematic factors. Second, given the additivity of idiosyncratic risk, an equity long portfolio with high

idiosyncratic risks necessarily over-weights stocks with high idiosyncratic variations. Such stocks are

shown by the literature (Durnev, Morck, Yeung, and Zarowin (2003) and Durnev, Morck, Yeung (2004))

as having a higher ratio of private, firm-specific information to noise, and carrying more information that

gets impounded into the price through informed trading (Chen, Goldstein, and Jiang (2006)). Finally, this

pattern is related to Agarwal, Fos, and Jiang’s (2009) finding that hedge funds which choose not to report

to any commercial databases have significantly lower factor loadings compare to funds that do. Both

their findings and ours indicate that institutional investors who adopt less conventional investment

strategies value privacy more—they are more likely to refrain from voluntary disclosures or to seek

exemptions from mandatory ones.

B. Stock Characteristics and Confidential Holdings

The next question in line is what types of stocks are more likely to be included in confidential

holdings. If the primary purpose of confidential filing is to conceal private information, then stocks in

such holdings are likely to be associated with information-sensitive events (such as M&As), more opaque

and subject to more information asymmetry among investors compared to stocks that are revealed in the

regular quarterly filings.

In the SEC guideline for amendment to 13F filings, “open risk arbitrage” and “block positioning”

are allowable reasons for the delay in disclosure. The event that best exemplifies both motives is merger

and acquisition. An M&A arbitrage is usually made upon an announced attempt of acquisition after

which the risk arbitrageurs bet on the completion of the deal and the convergence of the price to the

bidding price. Therefore, we use the indicator variable (M&A) for a stock that was an M&A target during

19

the one-year period ending in the portfolio quarter as a proxy for the M&A arbitrage motive of the

confidential filing.

Data on M&A transactions are retrieved from Securities Data Company (SDC), updated to the

end of 2007.29 A necessary condition for the classification of an M&A transaction is a sufficient change-

of-control. For this purpose, we exclude transactions classified as acquisitions of partial stakes, minority

squeeze-outs, buybacks, recapitalizations, and exchange offers. We also require that the bidder had a

stake below 50% before the transaction and a stake above 50% afterwards. Our final sample has 4,786

deals during the period of 1998-2007.

More generally, we use several variables that are firm-specific drivers of information asymmetry

including firm size, liquidity, distress risk, and analyst following. Extant literature indicates that greater

information asymmetry is associated with smaller stocks (Chari, Jagannathan, and Ofer (1988), Llorente,

Michaely, Saar, and Wang (2002)), illiquid stocks (Glosten and Milgrom (1985), Merton (1987),

Diamond and Verrecchia (1991), and Kim and Verrecchia (1994)), lesser analyst following (Brennan and

Subrahmanyam (1995), Hong, Lim, and Stein (2000), Chang, Dasgupta, and Hilary (2006)), and higher

probability of financial distress.

Market capitalization (Size) at the quarter-end is obtained from CRSP. Book-to-market ratios

(B/M) are recorded at year-end using data from CRSP and COMPUSTAT. We also include the market

(CRSP value-weighted) adjusted past 12-month return (Adj. Past Return) to control for momentum. We

resort to the Amihud (2002) illiquidity measure as the proxy for trading liquidity (Illiquidity). The

measure is constructed as the yearly average of the square root of |return|/(price ×volume), essentially an

empirical analogue to the inverse of Kyle’s (1985) lambda, or the inverse of market depth. We measure

analyst coverage of a firm by counting the number of analysts in the I/B/E/S database (available through

WRDS) that make at least one forecast or recommendation on the firm during the year (Analysts).

Finally, probability of financial distress is measured by the distance-to-default (DtD), which

refers to the number of standard deviation decreases in firm value before it drops to the face value of debt

(i.e., the firm is in default). This measure is motivated by Merton’s (1974) bond pricing model populated 29 This data was obtained from Edmans, Goldstein, and Jiang (2009). We thank the authors for sharing the data.

20

by Moody’s KMV, and is now a standard measure for default risk. We estimate distance-to-default for

each firm at each year end following the estimation procedure in Vassalou and Xing (2004). Because DtD

is a one-sided measure, we use a dummy variable for DtD to be smaller than 1.64 as an indicator for non-

negligible distress risk (i.e., the estimated probability of distress being 5% or higher).

Panel A of Table IV reports the summary statistics of stock-level variables discussed above, and

compares between those included only in the original filings and those only in the confidential filings.30

For the purpose of comparison, we discard amendment filings which we could not find the original filings

for the same institution-quarter pairs. Within the confidential filings, stocks that are part of the ongoing

acquisition could be quite different from those of the ongoing disposition. However, the confidential

filings do not explicitly state the acquisition/disposition purpose. For practical purpose, we adopt the

following algorithm to separate the two. For each stock in a confidential filing, we compare the position

(number of shares) to that of the same stock by the same institution at the previous quarter-end, and

classify acquisition (disposition) by net increase (decrease).31 In case of no change (5.4% of the sample),32

we drop the observation whenever acquisitions and dispositions are separately analyzed. According to

this algorithm, 65.0% of the positions in our confidential holdings sample are classified as acquisition

motivated while the remainder are dispositions.

[Insert Table IV here.]

It is worth noting that the unit of observation in Table IV is at the institution-quarter-holding

level, and sample of stocks in the holdings is very different from the universe of stocks because a stock-

quarter observation appears in the sample as many times as it appears in the holdings of all filings

institutions. As a result, stocks with high institutional ownership are over-represented in the pooled

sample of holdings. Moreover, the sample adopted by this table excludes original filings that are not

paired with a confidential filing. We adopt this sample selection criterion in order to facilitate extracting

30 Some stocks appear in both the original and amendment filings of an institution-quarter, where the amendment restates the number of shares in the portfolio. These stocks are not included in this analysis. 31 We adjust for the stock splits while computing for changes in holdings across different quarters. 32 This sample of unchanged positions happens to over-lap largely with positions in consecutive confidential filings by the same institution. Hence the majority of these positions are included in our sample when they first appear in the first of the series of confidential filings by the same institutions.

21

information relating a stock’s characteristics to the probability of its being included in a confidential

holding when it could have been in the original holdings. Quarterly dummies are incorporated in all

specifications of Table IV.

Panel A of Table IV shows that stocks in acquisition-motivated confidential holdings are smaller,

have higher book-to-market, lower trading liquidity, lower analyst coverage, higher distress risk, and

higher momentum, compared to the stocks in the original filings. Differences along these dimensions are

statistically significant at the 10% or less level and generally point towards greater information

asymmetry in the confidential holdings. Moreover, stocks in confidential holdings are far more likely to

have been recent targets in M&A announcements, a probability of 10.5% versus 3.2% for the original

filings, suggesting that M&A risk arbitrage is an important motive underlying confidential treatment.

The example of Stark Onshore master fund tabulated in Table I represents a more extreme example where

39 (70.9%) out of these 55 holdings were targets in M&A transactions during the year ending in the

portfolio quarter.

The disposition-motivated confidential holdings demonstrate some contrasts compared to their

acquisition-motivated counterparts. They are no smaller in market-cap, no less liquid, and have no less

analyst coverage than holdings in the original filings. On the other hand, the disposition-related

confidential holdings also have higher distress risk and a higher-than-normal probability involving stocks

that were M&A targets, but the incremental probability of M&A stocks is less than one-quarter the level

of the acquisition-related confidential holdings. Overall, the evidence is much weaker that disposition-

related confidential holdings are motivated by private information.

In addition to the univariate analyses, we explore the same issue using multivariate logistic

regressions. The model specification is as follows:

, , , , ,( 0),i j q i q q Ind i j qCH StockCharλ α δ ε= + + + > (8)

where , ,i j qCH is a dummy variable equal to one if stock i is in the confidential holdings of institution j in

quarter q. The all-sample average of , ,i j qCH is 6.6% (2.6%) for the acquisition- (disposition-) motivated

sample. ,i qStockChar is the same vector of stock characteristics variables used in Panel A of Table IV.

22

All standard errors in these regressions adjust for heteroskedasticity and clustering at the stock level. In

addition to the quarterly dummies ( qα ), the Fama-French 10 industry dummies ( Indδ ) are added to

regression (8) to control for unobserved heterogeneity at the industry level. Results without the industry

dummies are qualitatively similar and marginally stronger.

Reported in Table IV Panel B are the estimated coefficients λ̂ , their associated t-statistics, and the

average partial effects (APE) of the ,i qStockChar variables. More specifically, the APEs are computes as

the empirical analogue to ( ), , , ,Pr 1 / |i j q i q i qE CH StockChar StockChar⎡ ⎤∂ = ∂⎣ ⎦ :

( ) ( )( )( )

, ,, ,

,

2, ,

,

1ˆ ˆ ˆ ˆ ˆˆ ˆ1

ˆ ˆˆexp1ˆ ,ˆ ˆˆ1 exp

i q q Ind i q q Indi j q

i q q Ind

i j qi q q Ind

APE StockChar StockCharn

StockChar

n StockChar

λ λ α δ λ α δ

λ α δλ

λ α δ

⎡ ⎤= Λ + + − Λ + +⎣ ⎦

+ +=

⎡ ⎤+ + +⎣ ⎦

∑

∑ (9)

where ( )Λ i is the cumulative probability function for logistic distribution.

Again, the sample for regression (8) includes only holdings in paired original-confidential

filings. Using the full sample (including positions of original filings without paired confidential filings)

would also yields consistent results, but the power of the test would be lower due to the large number of

observations with very little information content because most of the unpaired original filings are made

by the great majority (more than 90%) of institutions that have never resorted to confidentiality.33

Moreover, a logistic regression has the desirable feature that all of its slope coefficients (but not the

intercept) have the same probability limit as those using the full sample, but the former are more efficient

estimates.

Results from multivariate logistic regressions, as reported in Table IV Panel B, provide messages

consistent to those from Panel A. Because Size, Analyst, and Illiquidity have high pairwise correlations

33 In general, a discrete response regression model suffers from low power if the unconditional probability of a positive response is miniscule (as would be the case if we use the full sample). In such cases, “choice-based sampling” such as eliminating observations that have a zero probability to have a positive response (such as holdings of institutions that never resort to confidential filing) can increase the power of the test by increasing the average information content of the kept observations, Please see Manski and McFadden (1981) for a general discussion of the approach.

23

(with absolute values above 0.60), we try specifications that have only one of the three at a time, as well

as having all three in one regression. Overall, acquisition-motivated confidential holdings are strongly

associated with characteristics that proxy for higher level of information, while the same relation is much

weakened for disposition-motivated confidential holdings. Moreover, confidential holdings include stocks

that are both past winners and losers conditional on other characteristics, but with a clear dichotomy:

while high past returns significantly increase the probability of a stock being included in the acquisition-

related confidential holdings, the opposite is true for disposition-related ones.

IV. Performance of portfolios of confidential filings

Having examined the determinants of confidential holdings and shown that such holdings seemed

to be motivated by private information or perception of private information; it is natural to ask whether

confidential holdings are associated with superior returns ex post.

A. Choice of Performance Measure

The default performance measure adopted in this paper is the Daniel, Grinblatt, Titman, and

Wermers (1997) holdings-based measure (henceforth the “DGTW measure” or “DGTW benchmark-

adjusted return”). More specifically, we form 125 portfolios, in June of each year, using all the common

stocks listed on NYSE, AMEX and NASDAQ based on a three way quintile sorting along the size (NYSE

size-quintile), book-market ratio, and momentum dimensions. The daily DGTW benchmark return for

each portfolio is the value-weighted return of all the component stocks. Finally, the abnormal

performance of a given stock is its return in excess of that of the benchmark portfolio it belongs.

The choice of the performance measure reflects the focus on superior returns associated with

stock picking. According to Daniel, Grinblatt, Titman, and Wermers (1997), the superior performance of

a money manager can be decomposed into three elements: stock selectivity, style timing, and execution

costs. Given that applications for confidential treatment need to be made at the individual stock level, the

justifiable private information should be stock-specific rather than about asset classes or overall market

timing. Further, our analyses are based on holdings that do not incorporate transaction costs. Therefore,

24

we use the DGTW measure which corresponds to the stock characteristic selectivity component.

A. Comparing Return Performance of Confidential and Original Holdings

We first compare the DGTW benchmark-adjusted returns of confidential (separately for

acquisition and disposition motivated) and original holdings at the institution-quarter level. This

procedure includes all paired original and confidential filings in the sample and entails four steps. First,

we separate all stocks in each confidential filing into acquisition- and disposition-motivated holdings.

Second, we compute the DGTW measure for each stock in each group (original holdings, confidential

holdings—acquisition, and confidential holdings—disposition). Third, we compute both the value-

weighted (Panel A of Table V) and equal-weighted (Panel B of Table V) average performance of

portfolios of each group. As a result, there are up to three portfolios for each institution-quarter. Finally,

we average, over all original and all acquisition- and disposition-motivated confidential portfolios, their

DGTW measures at different return horizons ranging from one to 12 months from the quarter-end

portfolio date. A 45-day horizon is added to reflect the normal delay allowed for the regular 13F filings.

[Insert Table V here.]

We conduct two-sample mean difference t-tests that compare the performance of the original

portfolios with the two types (acquisition and disposition) of confidential portfolios. The differences

adjust for the quarterly fixed effects, and the t-statistics are based on standard errors that are robust to

heteroskedasticity as well as clustering at the institution level.

Results in Table V provide evidence that acquisition-motivated confidential holdings exhibit

higher risk-adjusted returns compared to original holdings. The pairwise differences are positive for the

first six months, and are statistically significant at the 5% (10%) or less level for the first three (four)

months. Using the value-weight numbers, the difference peaks at 1.09% in two months (t-statistics =

3.44), equivalent to an annualized return spread of 6.54%. Therefore, positions that institutions

accumulate and choose to hide in confidential holdings seem to be more informed than the average

holdings in the same institutions’ overall portfolio, but the return spread peaks quickly, implying that the

private information is short-lived.

25

In contrast, the disposition-motivated confidential holdings do not exhibit any abnormal returns

over the original holdings at any time horizon. If institutions sell stocks based on superior information

and hide such selling using confidential filings, the superior information should manifest itself in

significant negative abnormal returns of these stocks after the quarter end. This is not the case, indicating

that confidential holdings related to on-going dispositions are more likely to be motivated by liquidity,

rather than by private information. In such cases, confidential filings may still benefit the institutions in

mitigating the adverse price impact that might ensue had the institution carried out the disposition in the

open. Not surprisingly, a direct comparison between acquisition- and disposition-motivated confidential

holdings by the same institutions (not tabulated) indicates that the former provides higher returns: The

difference peaks at 1.29% (1.39%) at the three month horizon using equal (value) weights (significant at

the 10% level).

In untabulated analyses, we conduct two sensitivity checks. First, we repeat the calculation in

Table V but separately for the subsample of confidential filings that result from SEC approvals of the

confidential treatment and those responding to SEC denials. Interestingly, the abnormal returns on the

two subsamples are statistically indistinguishable for all time horizons considered. The lack of difference

indicates that institutions’ ability or willingness to provide adequate factual support for their confidential

holdings is not indicative of the quality of their private information, neither is the SEC’s judgment on

whether an application for confidentiality merits approval.

Second, we replicate the analyses in Table V using the daily four-factor alpha (Carhart (1997)) as

the return performance measure. The resulting abnormal performance is comparable to that reported in

Table V using the DGTW measure. The value-weighted average daily alpha of the acquisition-motivated

confidential holdings peaks at 1.7 basis points at the three-month horizon (equivalent to an annualized

spread of 4.28%), significant at the 1% level. Moreover, the daily alpha measure remains strong and

significant (at the 5% level) at 1.1 basis points at the one-year horizon. Consistent with results in Table

V, no significant performance is detectable for the disposition subsample using the alpha measure.

B. Abnormal Returns of Confidential Holdings: Cross-Sectional Evidence

26

We next decompose the risk-adjusted performance of confidential holdings by both filing and

institution characteristics. For expositional convenience, we henceforth report the results for only the

equally-weighted case as both equally-weighted and value-weighted portfolios yield qualitatively similar

findings. In most cases, the value-weighted scheme produces slightly stronger results (as shown in Table

V). Based on the results in Tables IV and V and discussions in the previous sections, we focus on

acquisition-motivated confidential holdings only.

First, we group institutions by their types as defined in Table III and conduct the following

regression:

, , 1 2 , ,( ) ( )i j t i i t i j tDGTW j Conf InstType j Orig InstTypeβ β λ ε= = ⋅ + = ⋅ + + . (10)

In (10), the dependent variable, , ,i j tDGTW , is the equally-weighted DGTW benchmark-adjusted return of

the quarterly portfolios of institutions i of filing type j in quarter t. Filing type j takes two values: j=Conf

indicates that the filing is confidential and j = Orig indicates original filing. InstTypei is a vector of

institution types as defined in Table III. The interaction of filing type and institution type yields eight

dummy variables: The original and confidential filings by banks and insurance companies (BKORIG and

BKCONF); those of hedge funds (HFORIG, HFCONF), of investment companies/advisors (INVORIG,

INVCONF), and of other institutions (OTHORIG, OTHCONF). In the regression, we designate BKORIG

as the omitted category. Finally, tλ is the quarterly dummy variable. All standard errors are robust to

heteroskedasticity and are clustered at the filing institution level. Results are reported in the Panel A of

Table VI.

[Insert Table VI here.]

Panel A shows that the institution-filing categories exhibiting the highest abnormal returns are

from the confidential holdings of hedge funds (HFCONF) and investment companies/advisors

(INVCONF). The differences relative to the omitted category, peaking at 1.85% for hedge funds and

1.48% for investment companies, are significant up to 4-9 months after the quarter-end portfolio date.

Moreover, the original holdings of these two groups of institutions also exhibit significant abnormal

returns relative to the omitted category for most horizons up to a year. It is perhaps a puzzle that the

27

confidential holdings of the bank/insurance company group have worse (albeit insignificant) performance

relative to its own original holdings for most of the time horizons up to a year.

Panel B of Table VI reports two-sample t-tests on the comparison of original versus confidential

holdings within each type of institutions and the pair-wise comparison of confidential holdings of

different types of institutions. We also group hedge funds and investment companies/advisors to form a

broader category of institutions that specialize in asset management. The tests confirm that the

confidential holdings of asset management oriented institutions have superior returns relative to their own

original holdings, and relative to the confidential holdings of banks, insurance companies, and other

institutions.34

Second, we explore the cross-sectional variation in the abnormal returns of confidential holdings

at the stock-quarter level, where the stock characteristics variables are the same as used in Table IV.

Results are reported in Table VII. We find that stocks—conditional on being in confidential holdings—

with relatively high trading liquidity, high analyst coverage, low financial distress risk, and high book-to-

market ratio are associated with higher abnormal returns over most horizons from one month to one year.

And lower market cap and past returns are associated with higher abnormal returns in the short time

horizon (within three months). Interestingly, these characteristics, according to Table IV, would

unconditionally make a stock an unlikely candidate to be included in confidential holdings. The contrast

suggests that if a stock is a priori unlikely to be sought for confidential treatment, then there needs to be

stronger reason (private information) for the stock to be part of the confidential holdings, and hence

higher abnormal returns conditional on being included in confidential treatment.

[Insert Table VII here.]

C. Discussion of Abnormal Returns

Tables V to VII collectively show that abnormal performance of confidential holdings of asset

management oriented institutions (i.e., hedge funds and investment companies/advisors) is significantly

34 The “Other” category exhibits volatile results that lack any pattern. This is mostly due to the small sample size: This category accounts for less than 5% of all institutions and about 6% of all confidential filings.

28

better than that of their own original holdings and that of other institutions. However, the economic

magnitude of the abnormal returns is relatively modest, raising some questions regarding the value of

private information underlying the confidential treatment. In this section we discuss the economic

interpretation of the results based on both empirical evidence from this study and that in the literature on

return performance of portfolio managers and institutional investors.

First, our results should be taken into the context of the general lack of superior performance

among active portfolio managers as a whole. A large literature on the return performance of institutional

investors has yet to find economically significant and persistent superior performance of any large sample

of institutional investors. French (2008) shows that the typical investor in an actively managed portfolio

underperforms the market by 67 basis points during 1980-2006. Yet the great majority of the assets are in

the hands of active rather than passive (indexed) strategies.35 A large number of portfolio managers

believe (or make believe) that they have superior information to justify active management, but in

equilibrium their perceived private information cannot all materialize in superior returns because the

aggregate performance of active managers can only be at par with the market before fees. As a

comparison, we are able to identify a small subset of actively managed portfolios that are most likely to

contain private information and that indeed exhibit superior performance to some degree. It is still

possible that institutional investors perceive themselves as possessing more superior information than

truth (over-confidence might be one explanation for the inconsistent belief). They may seek confidential

treatment more frequently or for a larger collection of stocks than are warranted by genuinely valuable

information, resulting in modest overall performance of the confidential holdings.

Second, the low cost of seeking confidentiality should be reflected in equilibrium. The writing

and filing of confidentiality applications can to a large extent be standardized and specialized within an

institution; and even a frivolous application that leads to denial by the SEC still provides comfortable

delay in disclosure (see discussion in Section II.A.). Therefore, it is possible that institutions exhaust the

opportunities for delayed disclosure until the private information on the marginal cases becomes minimal.

35 Aggregate statistics are hard to obtain, however, the numbers from the mutual fund sector are informative. According to the CRSP Mutual Funds database, index funds account for 4.5% of all mutual funds, and manage 15.5% of all assets in 2006.

29

Alternatively, institutions could trade more aggressively on their private information than if confidential

treatment were not available. In equilibrium, as a result, the confidential holdings as a whole (which

contain many marginal filings) do not exhibit impressive abnormal returns on average.

Third, some of the motives underlying confidentiality are not necessarily related to superior

information or return prediction. For example, an institution may choose to hide an on-going acquisition

or disposition that is liquidity motivated so as to avoid being front run upon. In such cases, the counter-

factual (that is, returns in the absence of confidentiality) could be negative rather than zero on average.

Moreover, stock characteristics associated with low trading liquidity tend to overlap with those capturing

information asymmetry. And the liquidity story is also broadly consistent with results in Table IV.

Plausibly, disposition is more likely to be liquidity-driven than acquisitions, which explains the lower

abnormal returns of disposition-motivated confidential holdings relative to the acquisition-motivated (see

Table V). Another reason for seeking confidentiality by some institutions could be to hide “losers” as a

more cost effective way of “window dressing” compared to trading ahead of the quarter end. The

significantly negative coefficient on Adj. Past Return in the disposition subsample reported in Table IV

Panel B is consistent with this story. This motive does not have a clear prediction on the abnormal

performance of the hidden positions.

Fourth, some institutions may seek confidentiality in order to prevent an investment strategy from

being “reverse engineered” or to confuse potential copycats (such as the cases of Berkshire Hathaway and

D. E. Shaw illustrated in the Introduction and Section I). In other words, the motive behind some

confidential filings is to “blur” a trading strategy or an investment pattern rather than to hide specific

positions. This is similar to strategic trading by informed traders to avoid revealing private information

where some of the trades do not necessarily lead to superior returns on their own (Kyle (1989)). This

motivation, if justified, should relate confidentiality seeking to a higher performance of the overall

portfolio (including both original and confidential holdings), rather than incremental performance on the

part of the confidential holdings. We have some evidence supporting this hypothesis. In an untabulated

analysis that examines the relation between the total portfolio return of confidentiality-seeking institution-

quarters and the covariates used in Table VI, we find that higher portfolio turnover rates, lower loadings

30

on the four common factors, and more frequent use of confidential treatment at the institution level are

associated with significantly higher abnormal return at short horizons (within three months after the

quarter-end portfolio date). In other words, when institutions adopting more active management or less

conventional investment strategies seek confidential treatment, their overall portfolio returns are higher.

And frequent users of confidentiality also exhibit higher returns on their total portfolios. However, the

magnitude of all these effects is small in economic terms and does not explain a serious portion of the

cross-sectional variation.

Finally, the implicit assumption that portfolios are formed right at the quarter-end may have

biased down the return results if the positions are actually accumulated throughout the quarter. However,

this stringent assumption is necessary to avoid any look-back bias or attributing superior performance to

momentum trading, and is the default method adopted by the literature that analyzes returns using

holdings data. Nevertheless, we repeat the analysis in Table V by assuming the beginning of the quarter

as the portfolio formation date in order to gain some insights into this possibility. Untabulated results

show that the average value-weighted abnormal return of acquisition-related confidential holdings relative

to the original holdings of the same institution-quarter rises to 4.1% (significant at the 1% level) in three

months after the quarter-end, which quadrupled- the magnitude documented in Table V. Such a method

would also strengthen results in Table VI, where the performance of confidential holdings by hedge funds

and investment companies/advisors exceeds that of their original holdings by a larger margin if the

quarter beginning is used as the portfolio formation date. The truth is probably somewhere between, but

we do not wish to over-interpret the strengthened results given the possible look-back bias for any

assumed portfolio formation date other than the quarter-end.

V. Policy Implications and Concluding Remarks

By relegating some of the quarter-end holdings to an amendment 13F filing, an institution could

delay the disclosure of such holdings to the public for a significant period of time (usually up to one year).

To the extent that institutions resort to confidentiality strategically, holdings in the confidential filings are

likely to contain more private information than those disclosed at regular intervals. Despite their potential

31

importance, the confidential holdings have not been included in the extant research that analyzes quarterly

holdings of institutions due to data availability through conventional sources.

This study fills the gap by examining the complete holdings that incorporate information from

confidential filings. Our results show that asset management companies, who rely on collecting and

process private information in order to deliver superior returns, seek confidentiality more often; their

holdings are more likely to be associated with private information and to exhibit superior returns as a

result. These findings offer an explanation to the ongoing resistance by investment managers against

ownership disclosure.

Our study prompts several implications for the researchers and regulators. First and foremostly, it

is probably good news to researchers who have been using the regular 13F quarterly holdings database

(such as the one maintained by Thomson Reuters): the proportion of the “hidden” positions out of all

institutional portfolios is small unconditionally (typically between 1% and 1.5% during most of the years

in our sample period); and the return difference between the confidential and disclosed holdings is modest

and short-lived overall. Hence, the bias of ignoring confidential holdings is likely to be small if the

purpose of the research is to track aggregate institutional ownership in public companies or to assess the

overall portfolio performance of any large sample of institutional investors.

On the other hand, given the importance of confidential holdings conditional on a confidential

filing (on average 27.3% of the total value of an institution-quarter portfolio as shown in Table II Panel

B), their disproportionate association with information sensitive events (such as M&As), and their

characteristics that point to general information asymmetry, ignoring confidential holdings could be a

significant omission in analyzing position changes of individual institutions or in response to specific

events (such as M&A arbitrage).

On the regulation front, our study raises interesting questions regarding the design of ownership

disclosure rules that optimally balance between ensuring sufficient transparency and preserving the

incentives for sophisticated investors to collect and benefit from private information. Our findings

indicate that confidential treatment, on top of the regular 45-day delay in regular 13F filing, offers

adequate relief for institutions who wish to disclose holdings only after a significant delay. The fact that

32

over 90% of the institutions never resort to confidential treatment indicates that the demand for

confidentiality is not wide-spread.

On the other hand, the heavy concentration of confidentiality seeking among few institutions (the

top ten users account for 40% of all amendment filings) and the high rejection rates among some of the

heavy users (such as D. E. Shaw and Caxton) alert the possibility of abuse. At least for some period in

our sample period (which was already after the SEC’s tightening of the rule in 1998), there seems to be a

lack of serious cost for seeking confidential treatment as some institutions repeatedly sought to hide a

large portion of their portfolios (rather than a handful of positions). Though properly denied by the SEC

for lack of stock-specific factual support, these applications hardly enjoyed less benefit in delayed

disclosure compared to the approved cases given the turnaround time of SEC review (see discussion in

Section II.A). Such patterns call for attention to refine the rules so as to make the cost and benefit clear

and consistent to all players. Finally, cases like Berkshire Hathaway’s confidential filing of its Wells

Fargo position in 1987 (see discussion in Section I) highlights another subtle balance between protecting

some institutions’ legitimate private information and preventing them from misleading the investor public

by withholding material information.

33

References Agarwal, Vikas, Vyacheslav Fos, and Wei Jiang, 2009, Inferring reporting bias in hedge fund databases

from hedge fund equity holdings, Working Paper, Georgia State University and Columbia University.

Amihud, Yakov, 2002, Illiquidity and stock returns: cross-section and time-series effects, Journal of Financial Markets 5, 31-56.

Aragon, George O., and J. Spencer Martin, 2009, A unique view of hedge fund derivatives usage: Safeguard or speculation?, Working Paper, Arizona State University and Carnegie Mellon University.

Brennan, Michael J., and Avanidhar Subrahmanyam, 1995, Investment analysis and price formation in securities markets, Journal of Financial Economics 38, 361-381.

Brunnermeier, Markus K., and Stefan Nagel, 2004, Hedge funds and the technology bubble, Journal of Finance 59, 2013-2040.

Carhart, Mark M., 1997, On persistence in mutual fund performance, Journal of Finance 52, 57-82.

Chang, Xin, Sudipto Dasgupta, and Gilles Hilary, 2006, Analyst coverage and financing decisions, Journal of Finance 61, 3009-3048.

Chari, V. V., Ravi Jagannathan, and Aharon Ofer, 1988, Seasonalities in securities returns: The case of earnings announcements, Journal of Financial Economics 21, 101-121.

Chen, Qi, Itay Goldstein, and Wei Jiang, 2006, Price informativeness and investment sensitivity to price, Review of Financial Studies 20, 619-650.

Chen, Hsiu-Lang, Narasimhan Jegadeesh, and Russ Wermers, 2000, The value of active mutual fund management: An examination of the stockholdings and trades of fund managers, Journal of Financial and Quantitative Analysis 35, 343-368.

Daniel, Kent, Mark Grinblatt, Sheridan Titman, and Russ Wermers, 1997, Measuring mutual fund performance with characteristic-based benchmarks, Journal of Finance 52, 1035-1058.

Diamond, Douglas W., and Robert E. Verrecchia, 1991, Disclosure, liquidity, and the cost of capital, Journal of Finance 66, 1325-1355.

Durnev, Art, Randall Morck, and Bernard Yeung, 2004, ‘Value Enhancing Capital Budgeting and Firm-Specific Stock Return Variation, Journal of Finance, 59, 65–105.

Durnev, Art, Randall Morck, Bernard Yeung, and Paul Zarowin, 2003, Does Greater Firm-Specific Return Variation Mean More or Less Informed Stock Pricing, Journal of Accounting Research, 41, 797–836.

Edmans, Alex, Itay Goldstein, and Wei Jiang, 2009, Takeover Activity and Target Valuations: Feedback loops in financial markets, Working Paper, University of Pennsylvania and Columbia University.

Fama, Eugene F., and Kenneth R. French, 1997, Industry costs of equity, Journal of Financial Economics 43, 153-193.

Frank, Mary M., James M. Poterba, Douglas A. Shackelford, and John B. Shoven, 2004, Copycat funds: information disclosure regulation and the returns to active management in the mutual fund industry,

34

The Journal of Law and Economics 47, 515-541.

French, Kenneth, 2008, Presidential Address: The Cost of Active Investing, Journal of Finance 63, 1537-1573.

Ge, Weili, and Lu Zheng, 2006, The frequency of mutual fund portfolio disclosure, Working Paper, University of Washington and University of California, Irvine.

Glosten, Lawrence R., and Paul R. Milgrom, 1985, Bid, ask, and transaction prices in a specialist market with heterogeneously informed traders, Journal of Financial Economics 14, 71-100.

Griffin, John M. and Xu, Jin, 2009, How smart are the smart guys? A unique view from hedge fund stock holdings, Review of Financial Studies 22, 2531-2570.

Grinblatt, Mark, and Sheridan Titman, 1989, Mutual fund performance: An analysis of quarterly portfolio holdings, Journal of Business 62, 393-416.

Grinblatt, Mark, and Sheridan Titman, 1993, Performance measurement without benchmarks: an examination of mutual fund returns, Journal of Business 66, 47–68.

Grinblatt, Mark, Sheridan Titman, and Russ Wermers, 1995, Momentum investment strategies, portfolio performance, and herding: A study of mutual fund behavior, American Economic Review 85, 1088-1105.

Hong, Harrison, Terence Lim, and Jeremy C. Stein, 2000, Bad news travels slowly: Size, analyst coverage and the profitability of momentum strategies, Journal of Finance 55, 265-295.

Huang, Lixin, and Jayant R. Kale, 2009, The Effect of Supplier and Customer Industry Interrelations on Mutual Fund Investment and Performance, Working Paper, Georgia State University.

Kacperczyk, Marcin, Clemens Sialm, and Lu Zheng, 2005, On the industry concentration of actively managed equity mutual funds, Journal of Finance 60, 1983-2011.

Kacperczyk, Marcin, Clemens Sialm, and Lu Zheng, 2008, Unobserved actions of equity mutual funds, Review of Financial Studies 21, 2379-2416.

Kempf, Alexander, and Klaus Kreuzberg, 2004, Portfolio disclosure, portfolio selection, and mutual fund performance evaluation, Working Paper, University of Cologne.

Khandani, Amir E., and Andrew W. Lo, 2007, What happened to the quants in August 2007?, Journal of Investment Management 5, 29-78.

Kim, Oliver, and Robert E. Verrecchia, 1994, Market liquidity and volume around earnings announcements, Journal of Accounting and Economics 17, 41-68.

Kyle, Albert S., 1985, Continuous auctions and insider trading, Econometrica 53, 1315-1335.

Kyle, Albert S., 1989, Informed speculation with imperfect competition, Review of Economic Studies 56, 317-355.

Llorente, Guillermo, Roni Michaely, Gideon Saar, and Jiang Wang, 2002, Dynamic volume-return relation of individual stocks, Review of Financial Studies 15, 1005-1047.

Manski, Charles F., and Daniel McFadden, 1981, Alternative estimators and sample designs for discrete

35

analysis, in Charles F. Manski and Daniel McFadden ed.: Structural Analysis of Discrete Data with Econometric Applications (MIT Press), 2-50.

Merton, Robert C., 1987, A simple model of capital market equilibrium with incomplete information, Journal of Finance 42, 483-510.

Musto, David K., 1997, Portfolio disclosures and year-end price shifts, Journal of Finance 52, 1563-1588.

Musto, David K., 1999, Investment decisions depend on portfolio disclosures, Journal of Finance 54, 935-952.

Titman, Sheridan, and Christian Tiu, 2009, Do the best hedge funds hedge? Working paper, University of Texas, Austin.

Vassalou, Maria, and Yuhang Xing, 2004, Default risk in equity returns, Journal of Finance 59, 831-868.

Wermers, Russ, 2000, Mutual fund performance: An empirical decomposition into stock-picking talent, style, transaction costs, and expenses, Journal of Finance 55, 1655-1703.

Wermers, Russ, 2001, The potential effects of more frequent portfolio disclosure on mutual fund performance, Investment Company Institute Perspective 7, 1-12.

Wermers, Russ, 2003, Is money really “smart”? New evidence on the relation between mutual fund flows, manager behavior, and performance persistence, Working Paper, University of Maryland.

Wermers, Russ, 2006, Performance evaluation with portfolio holdings information, North American Journal of Economics and Finance 17, 207–230.

Wermers, Russ, Tong Yao, Jane Zhao, 2007, The investment value of mutual fund portfolio disclosure, Working Paper, University of Arizona and University of Maryland.

36

Figure 1 Time Series of the Numbers of Original and Confidential 13F Filings

This figure shows the quarterly time series of the number of the original filings (scaled to the left axis), that of the amendment 13F filings and those with approved confidential treatment (scaled to the right axis) in our final sample. Section II.A provided a detailed description of the construction of the sample.

0

15

30

45

60

75

90

‐

400 

800 

1,200 

1,600 

2,000 

2,400 

# Amen

dmen

t Filing

s

# Origina

l Filings

Quarterly Holdings Date

# of Original Filings (left axis)# of Amendment Filings (right axis)# Approved confidential treatment (right axis)

37

Table I

Confidential Holdings of Stark Onshore Management LLC

This table lists all the common stock confidential holdings reported in the 13F amendments filed by Stark Onshore Management LLC over the sample period 1999Q1-2007Q2. A confidential holding is defined as a position in an amendment filing that was unreported in the corresponding original 13F filing, or that was reported in the original filing with a different number of shares. “Issuer Name” is the name of the company issuing the common stock. “Shares” is the number of shares held by Stark Onshore on the portfolio date. “Portfolio Date” is the quarter-end date for which the portfolio holdings are reported. “Filing Date” is the date when the 13F amendment is filed. “Thomson Reuters” is an indicator variable for whether the holding is reported to the Thomson Reuters Ownership Database. “M&A Target” is an indicator variable for whether the issuer company was a target for merger and acquisition during four-quarter period ending in the portfolio quarter.

Issuer Name CUSIP Shares

Portfolio Date Filing Date

Thomson Reuters

M&A Target

Anthem Inc 94973V10 67,360 9/30/2004 2/14/2005 No No Cox Communications Inc 22404410 269,964 9/30/2004 2/14/2005 No No Metro-Goldwyn-Mayer Inc 59161010 60,000 9/30/2004 2/14/2005 No Yes Sears Holdings 81238710 390,800 12/31/2004 5/13/2005 No Yes Symantec Corp 87150310 161,650 12/31/2004 8/16/2005 No No Gold Fields Ltd 38059T10 73,277 3/31/2005 8/16/2005 No No Symantec Corp 87150310 161,650 3/31/2005 8/16/2005 No No Sungard Data Systems 86736310 1,557,250 3/31/2005 9/27/2005 No Yes Unocal Corp 91528910 393,650 3/31/2005 9/27/2005 No No MCI Communications Corp 55269110 2,103,850 3/31/2005 2/15/2006 No Yes Sungard Data Systems 86736310 1,557,250 6/30/2005 9/27/2005 No Yes Unocal Corp 91528910 393,650 6/30/2005 9/27/2005 No Yes Brookstone Inc 11453710 98,463 6/30/2005 10/7/2005 No Yes Infousa Inc New Com 45670G10 221,542 6/30/2005 10/7/2005 No Yes Metals Usa Inc 59132420 183,275 6/30/2005 10/7/2005 No Yes Cablevision Systems Corp 12686C10 281,250 6/30/2005 1/6/2006 No Yes Medicis Pharmaceutical 58469030 13,750 6/30/2005 1/6/2006 No No AT&T Corp 00195750 6,250 6/30/2005 2/15/2006 No Yes MCI Communications Corp 55269110 1,119,450 6/30/2005 2/15/2006 No Yes Gold Banc Corp Inc 37990710 555,203 9/30/2005 12/15/2005 No No AT&T Corp 00195750 6,250 9/30/2005 2/15/2006 No Yes Bei Technologies Inc 05538P10 46,200 9/30/2005 2/15/2006 No Yes Cablevision Systems Corp 12686C10 281,250 9/30/2005 2/15/2006 No Yes Chiron Corp 17004010 506,040 9/30/2005 2/15/2006 No Yes Hibernia Corp 42865610 525,000 9/30/2005 2/15/2006 No Yes MCI Communications Corp 55269110 1,119,450 9/30/2005 2/15/2006 No Yes

38

Issuer Name CUSIP Shares

Portfolio Date Filing Date

Thomson Reuters

M&A Target

Medicis Pharmaceutical 58469030 13,750 9/30/2005 2/15/2006 No No Metals Usa Inc 59132420 185,775 9/30/2005 2/15/2006 No Yes Petrokazakhstan Inc 71649P10 93,750 9/30/2005 2/15/2006 No No Guidant Corporation 40169810 61,650 9/30/2005 5/19/2006 No Yes Boston Scientific Corp 10113710 506,250 12/31/2005 5/19/2006 No No Guidant Corporation 40169810 397,011 12/31/2005 5/19/2006 No Yes Ipayment, Inc 46262E10 26,360 12/31/2005 5/19/2006 No Yes Independence Comm. Bank Corp 45341410 373,797 12/31/2005 6/5/2006 No Yes Albertson's Inc 01310410 392,240 3/31/2006 6/5/2006 No Yes Independence Comm. Bank Corp 45341410 13,677 3/31/2006 6/5/2006 No Yes Education Management Corp 28139T10 411,591 3/31/2006 8/15/2006 No Yes Thomas Nelson 64037610 75,360 3/31/2006 8/15/2006 No Yes Capital One Financial 14040H10 110,000 3/31/2006 11/20/2006 No No Engelhard Corp 29284510 72,800 3/31/2006 11/20/2006 No Yes Keyspan Corp 14040H10 396,780 3/31/2006 2/20/2007 No Yes Capital One Financial 14040H10 145,000 6/30/2006 11/20/2006 No No Commercial Capital Bancorp, Inc 20162L10 443,073 6/30/2006 11/20/2006 No Yes Exelon Corp 30161N10 783,500 6/30/2006 11/20/2006 No No Fisher Scientific Intl 33803220 116,080 6/30/2006 11/20/2006 No Yes Kinder Morgan Inc 49455P10 202,340 6/30/2006 11/20/2006 No Yes Nco Group Inc 62885810 407,999 6/30/2006 11/20/2006 No Yes Public Service Enterprise Group 74457310 730,774 6/30/2006 11/20/2006 No No Keyspan Corp 49337W10 540,040 6/30/2006 2/20/2007 No Yes Longview Fibre Co 54321310 40,000 6/30/2006 2/20/2007 No Yes Constellation Energy Group Inc 21037110 648,660 6/30/2006 5/3/2007 No Yes Northwestern Corp 66807430 175,832 6/30/2006 5/3/2007 No Yes Univision Communications Inc 91490610 1,298,435 6/30/2006 5/3/2007 No Yes Multi Fineline Electronix In 62541B10 933,653 3/31/2007 5/16/2007 No No Rouse Co 77927310 269,910 9/30/2004 11/25/2004 Yes Yes

39

Table II Summary Statistics of 13F Original and Confidential Filings

Panel A of the table reports the distribution of the delay (in number of days) between the quarter-end portfolio date and the filing date for all original and confidential 13F filings (the “preliminary sample”). In Panel B, we use the “final sample” that excludes observations with extreme delays, i.e., more than 180 days for the original filings, and confidential filings with less than 45-day or more than 1,505-day (4 years plus 45 days) delay. Panel B summarizes the number of filings, the number of institutions, the dollar value, the number of stocks, and the average stock ownership share in the final sample. The types of institutions (Bank and Insurance, Hedge Fund, Investment Company/Advisor, and Other) are defined in Section III.A. The statistics for the two types of holdings are reported separately, and those of the confidential holdings are compared to the combined portfolio of the confidential filings and their corresponding original holdings. Panel C reports the number of confidential filings and percent of rejected filings of the top ten institutions that seek confidential treatment and the top ten institutions that are most frequently denied of their requests for confidential treatment. The institution types “HF” and “INVCO” are abbreviations of “Hedge Fund” and “Investment Company/Advisor”. Panel A: Delay Period between Portfolio Date and Filing Date Original 13F Filings Total Delay (days) 0-30 31-45 46-60 61-180 > 180        

Number 12,694 34,885 5,640 1261 705 55,185

Percent 23.00% 63.21% 10.22% 2.29% 1.28%   Confidential 13F Filings

Delay (days) 0-30 31-45 46-60 61-180 181-410 411-775 776-1505 > 1,505  

Number 34 107 129 518 752 288 103 27 1,958 Percent 1.74% 5.46% 6.59% 26.46% 38.41% 14.71% 5.26% 1.38% Total                 57,143

40

Panel B: Summary Statistics of Original and Confidential Holdings by Institution Types

Institution Type

Bank & Insurance

Hedge Fund

Investment Company/ Advisor

Other Total

Original 13F Filings # of institutions 346 950 1,827 123 3,246 # of 13F filings 6,203 14,265 31,749 1,937 54,154 $ million per institution-quarter (Mean) 6,690.0 1,509.2 3,176.4 3,986.6 3,168.7 $ million per institution-quarter (Median) 482.8 258.5 259.7 209.4 274.7 # of stocks per institution-quarter (mean) 411.6 119.9 166.5 264.7 185.8# of stocks per institution-quarter (median) 176.0 47.0 74.0 69.0 72.0 % of outstanding shares (Mean) 0.53% 1.12% 0.50% 1.03% 0.69% % of outstanding shares (Median) 0.06% 0.35% 0.10% 0.09% 0.13% Confidential 13F Filings # of institutions 21 104 98 10 233 # of 13F filings 52 779 660 101 1,592 $ million per institution-quarter (Mean) 2,350.0 691.7 779.7 145.9 781.2 % of original filing and conf. filing combined 11.6% 33.1% 19.4% 49.8% 27.3% $ million per institution-quarter (Median) 30.9 125.7 137.4 82.0 121.9 % of original filing and conf. filing combined 0.1% 24.4% 4.8% 59.8% 12.3% # of stocks per institution-quarter (mean) 117.9 77.0 60.2 13.0 69.0 % of original filing and conf. filing combined 11.9% 26.3% 14.3% 49.0% 21.8% # of stocks per institution-quarter (median) 7.0 6.0 11.0 12.0 8.0 % of original filing and conf. filing combined 0.5% 18.3% 3.5% 60.6% 9.4% % of outstanding shares (Mean) 0.92% 1.13% 1.20% 1.62% 1.17% % of outstanding shares (Median) 0.39% 0.67% 0.57% 0.63% 0.61%

41

Panel C: Top Ten Confidential Filers and Top Ten Denied Institutions

Top Ten Confidential Filers Inst. Type # Conf. Filings % Rejected

Chesapeake Partners Management Co. INVCO 108 6.5% Mineworkers Pen. Scheme OTHER 81 0.0% UBS O’Connor LLC HF 76 1.3% Satellite Asset Management HF 63 9.5% Berkshire Hathaway Inc INVCO 62 71.0% Price T. Rowe Assoc Inc INVCO 61 6.6% HBK Investments LP HF 49 26.5% Lehman Brothers Inc. INVCO 49 0.0% Stark Offshore Management, LLC HF 40 0.0% Polygon Investment Partners HF 39 0.0% Total 628 75 % of the full sample 39.4% 29.5%

Top Ten Institutions with Denied Confidential Requests Inst. Type # Conf. Filings % Rejected

Berkshire Hathaway Inc INVCO 62 71.0% D. E. Shaw & Co., Inc. HF 17 100.0% Relational Investors, L.L.C. HF 22 63.6% HBK Investments LP HF 49 26.5% Staro Asset Management, L.L.C. HF 34 38.2% Atlantic Investment Co INVCO 12 91.7% SAB Capital Advisors LLC HF 23 39.1% Caxton Corporation HF 9 100.0% Two Sigma Investments, LLC HF 10 80.0% Redsky Partners LLC HF 8 100.0% Total 246 146 % of the full sample 15.5% 57.5%

42

Table III Determinants of 13F Confidential Holdings—Institution Level

This table reports the results from probit and tobit regressions modelling the determinants of 13F confidential filings at the institution level. In columns (1) and (2), the dependent variable of the probit model is an indicator variable for a filing to be confidential. In columns (3) and (4), the dependent variable of the tobit model is the dollar value of confidential holdings as a percentage of the total dollar value of holdings (both original and confidential) for an institution-quarter. Reported are coefficient estimates, and their t-statistics (in parentheses) and associated average partial effects (APE, in percentage points). The types of institutions (Bank and insurance, hedge fund, investment company/advisor, and other) are defined in Section III.A. In columns (1) and (3), “Banks and insurance companies” serves as the omitted category. “Log(Age)” is natural logarithm of the number of years since the institution’s first appearance on Thomson Reuters. “PortSize” is the total equity portfolio size of an institution calculated as the market value of its quarter-end holdings. “Turnover” is the inter-quarter portfolio turnover rate calculated as the lesser of purchases and sales divided by the average portfolio size of the last and the current quarter. “PortHHI” is the Herfindahl index of the portfolio, calculated from the market value of each component stock. “PortRet” and “PortVol” are the monthly average return and volatility on the portfolio during the quarter, assuming that the institution maintains the holdings of the last quarter-end. “Flow” is defined as the increase in total portfolio value between two consecutive quarters net of the increase due to returns, expressed as a percentage of the portfolio size at the previous quarter-end. “BetaMkt”, “BetaSMB”, “BetaHML”, and “BetaMom” are the loadings on the Fama-French three factors (market, size, book-to-market) and the momentum factor using imputed monthly returns for the 36-month period ending in the current quarter, assuming that the institution always maintains the most recent past quarter-end holdings. Quarterly dummies are incorporated in all regressions. Standard errors adjust for heteroskedasticity and clustering at the institution level. Coefficients marked with ***, **, and * are significant at the 1%, 5%, and 10% level respectively

Dep. Variable: Indicator for Confidential Filing

Dep. Variable: Percent of Holdings Filed as Confidential

(1) (2) (3) (4) Hedge Fund 0.667*** 0.785***

(4.25) (4.35) 3.04% 1.52%

Investment Company/Advisor 0.245 0.308* (1.59) (1.84) 1.12% 0.59%

Other 0.425 0.531 (1.44) (1.53) 1.94% 1.03%

Log(Age) -0.085 -0.059 (-1.19) (-0.97) -0.32% -0.10%

Log(PortSize) 0.214*** 0.167*** (5.41) (5.73) 0.81% 0.29%

43

Dep. Variable: Indicator for Confidential Filing

Dep. Variable: Percent of Holdings Filed as Confidential

(1) (2) (3) (4) Turnover 2.494*** 2.212***

(6.28) (5.90) 9.39% 3.78%

PortHHI 1.778*** 1.559*** (6.20) (6.31) 6.69% 2.66%

PortRet -0.224 -0.064 (-0.28) (-0.09) -0.84% -0.1094%

PortVol 3.869** 3.716** (2.26) (2.40)

14.56% 6.35% Flow 0.027 0.030

(0.85) (1.10) 0.10% 0.05%

BetaMkt -0.285** -0.272** (-2.05) (-2.07) -1.07% -0.46%

BetaSMB -0.133 -0.120 (-0.87) (-0.91) -0.50% -0.21%

BetaHML 0.094 0.082 (1.07) (1.03) 0.35% 0.14%

BetaMom -0.409** -0.339** (-2.39) (-2.24) -1.54% -0.58%

Constant -2.344*** -3.904*** -2.651*** -3.291*** (-15.61) (-10.89) (-9.81) (-8.93)

Observations 54371 43902 54371 43902 Pseudo R-squared 0.04 0.14 0.04 0.14

44

Table IV Stock Characteristics of the Original and Confidential 13F Holdings

Panel A compares the summary statistics (mean, median, and standard deviation) of the

characteristics of stocks in original and confidential 13F holdings. Confidential holdings are classified as acquisition- (disposition-) motivated if the position shows a net increase (decrease) over the previous quarter-end. All variables, unless otherwise specified, are calculated at the fiscal year-end before the portfolio dates. “Size” is the quarter-end market capitalization of the stock in millions of dollars. “B/M” is the firm’s book-to-market ratio. “Adj. Past Return” is the stock return during the 12 months prior to the quarter-end portfolio date adjusted by CRSP value-weighted market return. “Illiquidity” is the Amihud (2002) illiquidity measure, or the yearly average of the square root of daily |Return|/(Price×Vol). “Analysts” is the number of I/B/E/S analysts covering the firm during the year. “DTD < 1.64” is the dummy variable for the Merton (1987) distance-to-default measure to be smaller than 1.64 (implying a 5% or higher default probability). “M&A” is an indicator variable that takes a value of 1 for the stock of the firm that was an M&A target during the four-quarter period ending in the portfolio quarter. The standard errors of the two sample t-tests adjust for clustering at stock and quarter level. Panel B reports the results from a logistic regression modelling the determinants of 13F confidential holdings at the stock level. The dependent variable is an indicator variable for a stock to be included in the confidential holdings of an institution-quarter. “Log(Size)” is the natural logarithm of “Size” and “Log(Analysts)” is the natural logarithm of one plus the “Analysts” measure defined above. The first (last) four columns examine separately the determinants of acquisition (disposition) -motivated confidential holdings. Each column reports estimated coefficients, their t-statistics (in parenthesis), and the average partial effects (APE, in percentage points). Quarterly dummies and Fama-French 10-industry dummies are included in all specifications in Panel B. All standard errors adjust for heteroskedasticity and clustering at the stock level. Coefficients marked with ***, **, and * are significant at the 1%, 5%, and 10% level respectively.

45

Panel A: Summary Statistics of Stock Characteristics of Original and Confidential Holdings

Size B/M Adj. Past Return Illiquidity Analysts

DTD < 1.64 M&A

Original 13F Form Filings Mean 7,692.8 0.55 12.9% 0.12 12.31 20.1% 3.2% Median 1,300.8 0.44 1.5% 0.05 10.00 Std. Dev. 20,639.9 0.44 64.0% 0.20 10.60 40.0% 17.6% # obs 370,141 370,132 370,141 368,655 369,528 370,141 370,141

Confidential 13F Form Filings: Acquisition Sample Mean 5,510.2 0.60 17.7% 0.16 11.12 27.7% 10.5% Median 932.8 0.47 5.0% 0.07 8.00 Std. Dev. 16,299.3 0.51 69.7% 0.25 10.51 44.8% 30.7% # obs 26,127 26,125 26,127 26,047 26,086 26,127 26,127 Two-sample Tests Differences in Mean (Acquisition - Original) −2182.6*** 0.05* 4.8%* 0.04* −1.20** 7.7%*** 7.4%*** Clustered t-stat. (−3.60) (1.83) (1.79) (1.79) (−2.10) (2.90) (5.31)

Confidential 13F Form Filings: Disposition Sample Mean 6,949.6 0.59 11.2% 0.10 12.64 25.6% 4.7% Median 1,162.9 0.47 2.6% 0.06 10.00 Std. Dev. 20,001.4 0.47 58.0% 0.16 10.88 43.6% 21.2% # obs 9,906 9,905 9,906 9,872 9,888 9,906 9,906 Two-sample Tests Differences in Mean (Disposition - Original) −743.2 0.04*** −1.7% −0.02 0.32 5.5%*** 1.5%*** Clustered t-stat. (−1.00) (2.71) (−0.59) (−1.45) (0.59) (3.08) (2.95)

46

Panel B: Determinants of Confidential Holdings – Stock Level

Acquisition Sample    Disposition Sample

(1) (2) (3) (4) (5) (6) (7) (8) M&A 1.526*** 1.521*** 1.530*** 1.530*** 0.668*** 0.675*** 0.672*** 0.675***

(36.51) (36.40) (36.55) (36.54) (10.71) (10.75) (10.76) (10.77) 8.65% 8.62% 8.66% 8.66% 1.64% 1.65% 1.65% 1.65%

Log(Size) -0.105*** -0.092*** -0.015** -0.087*** (-21.17) (-13.97) (-2.04) (-8.34) -0.59% -0.52% -0.04% -0.21%

Illiquidity 0.502*** 0.063 -0.686*** -1.244*** (15.59) (1.50) (-9.13) (-11.21) 2.84% 0.36% -1.68% -3.05%

Log(Analysts) -0.106*** -0.021** 0.008 -0.011 (-14.18) (-2.26) (0.68) (-0.78) -0.60% -0.12% 0.02% -0.03%

DTD < 1.64 0.065*** 0.125*** 0.142*** 0.065*** 0.006 0.068** 0.024 0.015 (3.76) (7.27) (8.26) (3.80) (0.22) (2.33) (0.83) (0.51) 0.37% 0.71% 0.80% 0.37% 0.01% 0.17% 0.06% 0.04%

B/M -0.016 0.042** 0.069*** -0.019 -0.044 0.052** -0.023 -0.005 (-0.92) (2.41) (3.97) (-1.04) (-1.62) (1.98) (-0.86) (-0.18) -0.09% 0.24% 0.39% -0.11% -0.11% 0.13% -0.06% -0.01%

Adj. Past Return 0.077*** 0.032*** 0.040*** 0.069*** -0.155*** -0.126*** -0.156*** -0.085*** (7.74) (3.16) (3.90) (6.78) (-7.56) (-5.93) (-7.53) (-4.04) 0.44% 0.18% 0.23% 0.39% -0.38% -0.31% -0.38% -0.21%

Constant −1.602*** −2.498*** −2.218*** −1.665*** −4.659*** −4.719*** −4.795*** −3.924*** (−24.22) (−45.73) (−39.35) (−22.57) (−25.82) (−26.80) (−26.94) (−20.90)

Observations 396257 394691 395603 394691 380037 378517 379406 378517 Unconditional Mean 6.59% 6.60% 6.59% 6.60% 2.61% 2.61% 2.61% 2.61% Pseudo R-squared 0.12 0.12 0.12 0.12 0.12 0.12 0.12 0.12

47

Table V DGTW Benchmark-Adjusted Returns of Original and Confidential Holdings

This table reports the Daniel, Grinblatt, Titman, and Wermers (1997) (DGTW) benchmark-adjusted returns for original and confidential 13F holdings for return horizons ranging from one month to 12 months from the quarter-end portfolio date. A confidential holding is classified as acquisition-(disposition-) motivated if the position shows a net increase (decrease) over the previous quarter-end. The unit of observation is a portfolio that consists of all the original holdings, all the acquisition-motivated confidential holdings, or all disposition-motivated confidential holdings of an institution-quarter. The DGTW benchmark-adjusted returns are first computed for each stock in each portfolio and then are averaged at the portfolio level using value weights (Panel A) or equal weights (Panel B). For the two-sample mean difference t-tests, standard errors adjust for heteroskedasticity and clustering at the institution level, and the quarter dummies are added. Coefficients marked with ***, **, and * are significant at the 1%, 5%, and 10% level respectively.

Panel A: Value-Weighted DGTW Benchmark-Adjusted Returns

#Obs

Return Horizons

1m 45days 2m 3m 4m 5m 6m 9m 12m Confidential 13F Form Filings: Acquisition Sample Confidential Holdings 1,000 0.41% 0.47% 0.96% 1.10% 0.61% 0.74% 0.29% 0.01% 0.25% Original Holdings 54,144 −0.23% −0.23% −0.13% 0.03% −0.21% −0.06% 0.09% 0.19% 0.30% Diff.: Confidential − Original 0.64%*** 0.70%** 1.09%*** 1.07%*** 0.82%* 0.80% 0.20% −0.18% −0.05% Clustered t-stat. 2.95 2.55 3.44 2.87 1.85 1.39 0.34 −0.21 −0.01 Confidential 13F Form Filings: Disposition Sample Confidential Holdings 612 −0.73% −0.20% 0.62% 1.08% 0.28% 0.71% 0.77% 0.84% 0.69% Original Holdings 54,144 −0.23% −0.23% −0.13% 0.03% −0.21% −0.06% 0.09% 0.19% 0.30% Diff.: Confidential − Original −0.50% 0.03% 0.75% 1.05% 0.49% 0.77% 0.68% 0.65% 0.39% Clustered t-stat. −0.82 0.10 0.98 1.36 0.56 0.77 0.65 0.49 0.30

Panel B: Equally-Weighted DGTW Benchmark-Adjusted Returns

#Obs Return Horizons

1m 45days 2m 3m 4m 5m 6m 9m 12m Confidential 13F Form Filings: Acquisition Sample Confidential Holdings 1,000 0.49% 0.53% 1.04% 0.98% 0.48% 0.61% 0.35% −0.04% −0.11%

Original Holdings 54,154 −0.09% −0.07% 0.10% 0.17% −0.06% 0.16% 0.26% 0.43% 0.65%

Diff.: Confidential – Original 0.58%*** 0.60%** 0.94%*** 0.81%** 0.54% 0.45% 0.09% −0.47% −0.76%

Clustered t−Stat. 2.77 2.41 3.11 2.26 1.26 0.68 0.08 −0.67 -0.75

Confidential 13F Form Filings: Disposition Sample Confidential Holdings 612 −0.60% −0.30% 0.46% 0.51% −0.46% −0.23% −0.36% −0.70% −0.08%

Original Holdings 54,154 −0.09% −0.07% 0.10% 0.17% −0.06% 0.16% 0.27% 0.43% 0.65%

Diff.: Confidential − Original −0.51% −0.23% 0.36% 0.34% −0.40% −0.39% −0.63% −1.13% −0.73%

Clustered t-Stat. −0.98 −0.38 0.49 0.42 −0.44 −0.54 −0.70 −0.88 −0.37

48

Table VI Abnormal Returns of Original and Confidential Holdings and Institution Types

This table reports the results of multivariate regressions that examine the attribution of risk-adjusted performance of quarterly holdings to institutional characteristics and filing types. The dependent variable is the equally-weighted DGTW benchmark-adjusted returns (in percentage points) for return horizons ranging from one month to 12 months from the quarter-end portfolio date. Panel A reports the results of regressions of abnormal returns on institution types. The unit of observation is a portfolio that consists of all the original holdings or all the acquisition-motivated confidential holdings of an institution-quarter. The original holdings of the Bank/Insurance Company group serve as the omitted category. “BKCONF” is the indicator for confidential holdings of the Bank/Insurance Company category. “HFORIG” and “HFCONF” are the indicators of original and confidential holdings of the Hedge Fund category. “INVORIG” and “INVCONF” are the indicators of original and confidential holdings of the Investment Company/Advisor category. “OTHORIG” and “OTHCONF” are the indicators of original and confidential holdings of others types of 13F institutions. In Panel A, standard errors adjust for heteroskedasticity and clustering at the institution level and quarterly dummies are included. T-statistics are reported below coefficient estimates in parentheses. Panel B reports the results of pairwise comparisons of abnormal returns within/across different institution types. The differences of the coefficients are reported for each comparison, and asterisks are used to mark the significance. Coefficients or differences marked with ***, **, and * are significant at the 1%, 5%, and 10% level respectively.

49

Panel A: Regressions of Abnormal Returns on Institution Types

Return Horizon 1m 45days 2m 3m 4m 5m 6m 9m 12m HFORIG 0.057 0.060 0.355*** 0.543*** 0.531*** 0.784*** 0.864*** 1.288*** 1.688***

(1.14) (1.00) (4.78) (6.19) (4.68) (5.28) (5.48) (5.64) (5.68) HFCONF 0.674* 0.740* 1.149*** 1.851*** 1.375** 1.530*** 1.285** 1.439* 2.308

(1.89) (1.91) (3.10) (3.73) (2.30) (2.69) (2.00) (1.66) (1.64) INVORIG 0.004 0.034 0.161*** 0.249*** 0.222*** 0.368*** 0.401*** 0.588*** 0.656***

(0.12) (0.78) (2.75) (3.84) (2.80) (3.21) (3.42) (3.47) (2.95) INVCONF 0.896*** 0.871* 1.424*** 1.402** 1.479* 1.318 0.723 0.003 0.463

(2.71) (1.90) (2.71) (2.33) (1.95) (1.22) (0.69) (0.00) (0.26) OTHORIG -0.053 -0.270 0.066 0.050 -0.181 -0.237 -0.184 0.016 0.397

(-0.36) (-1.61) (0.38) (0.28) (-0.69) (-0.75) (-0.51) (0.03) (0.62) OTHCONF -0.020 0.701 1.941 -0.927 -1.145 2.083 -1.890 -3.592 -5.725

(-0.02) (0.73) (1.47) (-0.37) (-0.33) (0.53) (-0.26) (-0.49) (-0.58) BKCONF -0.242 -0.462 0.189 -2.398 -3.060 -4.304 -4.534 -1.276 -6.393

(-0.23) (-0.52) (0.08) (-1.46) (-1.27) (-1.41) (-1.36) (-0.32) (-1.31) Constant -0.247*** -0.261*** -0.319*** -0.260*** -0.476*** -0.472*** -0.365*** -0.494*** -0.541***

(-10.44) (-7.66) (-6.54) (-5.04) (-7.74) (-5.09) (-4.04) (-3.74) (-3.09) Observations 55143 55143 55143 55143 55143 55143 55143 55143 55143 R-squared 0.05 0.03 0.03 0.02 0.02 0.03 0.03 0.02 0.02

Panel B: Pairwise Comparisons within/across Different Institution Types

Return Horizon 1m 45days 2m 3m 4m 5m 6m 9m 12m Differences between Original and Confidential Holdings within Each Institution Type HFCONF−HFORIG 0.62%* 0.68%* 0.79%** 1.31%*** 0.84% 0.75% 0.42% 0.15% 0.62% INVCONF−INVORIG 0.89%*** 0.84%* 1.26%** 1.15%* 1.26%* 0.95% 0.32% −0.59% −0.19% (HFCONF+INVCONF)-(HFORIG+INVORIG) 0.76%*** 0.76%** 1.06%*** 1.30%*** 1.11%** 0.94% 0.48% -0.03% 0.48%

OTHCONF−OTHORIG 0.03% 0.97% 1.88% -0.98% -0.96% 2.32% -1.71% -3.61% -6.12%

Differences between Confidential Holdings across Different Institution Types HFCONF−INVCONF -0.22% -0.13% -0.28% 0.45% -0.10% 0.21% 0.56% 1.44% 1.85% HFCONF-BKCONF 0.92% 1.20% 0.96% 4.25%** 4.44%* 5.83%* 5.82%* 2.72% 8.70%* INVCONF-BKCONF 1.14% 1.33% 1.24% 3.80%** 4.54%* 5.62%* 5.26% 1.28% 6.86% HFCONF-OTHCONF 0.69% 0.04% -0.79% 2.78% 2.52% -0.55% 3.18% 5.03% 8.03% INVCONF-OTHCONF 0.92% 0.17% -0.52% 2.33% 2.62% -0.77% 2.61% 3.60% 6.19% (HFCONF+INVCONF)-(BKCONF+OTHCONF) 0.91% 0.71% 0.25% 3.33%** 3.57%* 2.69% 4.29% 3.14% 7.50%

50

Table VII

Abnormal Returns of Confidential Holdings and Stock Characteristics

This table reports the results of multivariate regressions that examine the attribution of risk-adjusted performance of acquisition-motivated confidential holdings to stock characteristics of the holdings. The dependent variable is the equally-weighted DGTW benchmark-adjusted returns for return horizons ranging from one month to 12 months from the quarter-end portfolio date. The unit of observation is a portfolio that consists of all the acquisition-motivated confidential holdings of an institution-quarter. The stock characteristic variables are the same as defined in Table IV. Standard errors adjust for heteroskedasticity and clustering at the stock level. Both quarterly and Fama-French 10-industry dummies are included. T-statistics are reported below coefficient estimates in parentheses. Coefficients marked with ***, **, and * are significant at the 1%, 5%, and 10% level respectively.

Return Horizon 1m 45days 2m 3m 4m 5m 6m 9m 12m Log(Size) -0.004*** -0.005*** -0.005*** -0.003** -0.003** -0.002 -0.003 -0.003 -0.003

(-4.81) (-4.68) (-3.93) (-2.09) (-2.05) (-1.26) (-1.27) (-0.91) (-0.68) Illiquidity -0.017** -0.020** -0.017 -0.030*** -0.031** -0.040** -0.045*** -0.076*** -0.089***

(-2.20) (-2.22) (-1.63) (-2.59) (-2.31) (-2.57) (-2.76) (-3.27) (-3.16) Log(Analysts) 0.001 0.005*** 0.007*** 0.005** 0.007*** 0.011*** 0.010*** 0.013*** 0.018***

(0.84) (3.20) (4.14) (2.56) (3.11) (4.09) (3.19) (3.06) (3.31) DTD<1.64 0.002 -0.004 -0.005 -0.010** -0.018*** -0.018*** -0.027*** -0.037*** -0.056***

(0.56) (-1.01) (-1.38) (-2.10) (-3.26) (-2.76) (-3.72) (-3.59) (-4.43) B/M 0.036*** 0.032*** 0.021** 0.013 0.010 -0.011 -0.002 -0.011 -0.024

(4.87) (3.63) (2.19) (1.25) (0.80) (-0.75) (-0.10) (-0.44) (-0.80) Adj. Past Return -0.013*** -0.012*** -0.010*** -0.006** -0.005* 0.003 0.001 -0.000 -0.008

(-7.62) (-5.99) (-4.37) (-2.20) (-1.65) (0.74) (0.20) (-0.06) (-1.26) Constant 0.036*** 0.032*** 0.021** 0.013 0.010 -0.011 -0.002 -0.011 -0.024

(4.87) (3.63) (2.19) (1.25) (0.80) (-0.75) (-0.10) (-0.44) (-0.80) Observations 51917 51917 51917 51917 51917 51917 51917 51917 51917 R-squared 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01